Session ID: ses_493bfadd3ffeZLgB0sjQrbNtGy Created: 12/29/2025, 5:35:13 PM Updated: 12/29/2025, 6:37:15 PM
Ah, I understand now! The issue is that the same sdk.chat is not a function error occurs when using Cloudflare AI Gateway as a provider. Let me investigate the Cloudflare provider implementation.
Tool: grep
**Input:**
```json
{
"pattern": "cloudflare",
"path": "/Users/englishm/git/opencode/packages/opencode/src"
}
Output:
Found 2 matches
/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts:
Line 356: "cloudflare-ai-gateway": async (input) => {
Line 377: baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
---
## Assistant
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 350,
"limit": 50
}Output:
<file>
00351| "X-Title": "opencode",
00352| },
00353| },
00354| }
00355| },
00356| "cloudflare-ai-gateway": async (input) => {
00357| const accountId = Env.get("CLOUDFLARE_ACCOUNT_ID")
00358| const gateway = Env.get("CLOUDFLARE_GATEWAY_ID")
00359|
00360| if (!accountId || !gateway) return { autoload: false }
00361|
00362| // Get API token from env or auth prompt
00363| const apiToken = await (async () => {
00364| const envToken = Env.get("CLOUDFLARE_API_TOKEN")
00365| if (envToken) return envToken
00366| const auth = await Auth.get(input.id)
00367| if (auth?.type === "api") return auth.key
00368| return undefined
00369| })()
00370|
00371| return {
00372| autoload: true,
00373| async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
00374| return sdk.chat(modelID)
00375| },
00376| options: {
00377| baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
00378| headers: {
00379| // Cloudflare AI Gateway uses cf-aig-authorization for authenticated gateways
00380| // This enables Unified Billing where Cloudflare handles upstream provider auth
00381| ...(apiToken ? { "cf-aig-authorization": `Bearer ${apiToken}` } : {}),
00382| "HTTP-Referer": "https://opencode.ai/",
00383| "X-Title": "opencode",
00384| },
00385| // Custom fetch to strip Authorization header - AI Gateway uses cf-aig-authorization instead
00386| // Sending Authorization header with invalid value causes auth errors
00387| fetch: async (input: RequestInfo | URL, init?: RequestInit) => {
00388| const headers = new Headers(init?.headers)
00389| headers.delete("Authorization")
00390| return fetch(input, { ...init, headers })
00391| },
00392| },
00393| }
00394| },
00395| cerebras: async () => {
00396| return {
00397| autoload: false,
00398| options: {
00399| headers: {
00400| "X-Cerebras-3rd-Party-Integration": "opencode",
(File has more lines. Use 'offset' parameter to read beyond line 400)
</file>
---
## Assistant
I see the issue! The `cloudflare-ai-gateway` provider has a `getModel` function that calls `sdk.chat(modelID)`, but similar to the GitHub Copilot Enterprise issue, it likely doesn't have the correct `npm` package specified, so the SDK isn't properly loaded.
Let me check how other providers handle this and where the npm package is set:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 775,
"limit": 50
}Output:
<file>
00776| if (provider.options) partial.options = provider.options
00777| mergeProvider(providerID, partial)
00778| }
00779|
00780| for (const [providerID, provider] of Object.entries(providers)) {
00781| if (!isProviderAllowed(providerID)) {
00782| delete providers[providerID]
00783| continue
00784| }
00785|
00786| if (providerID === "github-copilot" || providerID === "github-copilot-enterprise") {
00787| provider.models = mapValues(provider.models, (model) => ({
00788| ...model,
00789| api: {
00790| ...model.api,
00791| npm: "@ai-sdk/github-copilot",
00792| },
00793| }))
00794| }
00795|
00796| const configProvider = config.provider?.[providerID]
00797|
00798| for (const [modelID, model] of Object.entries(provider.models)) {
00799| model.api.id = model.api.id ?? model.id ?? modelID
00800| if (modelID === "gpt-5-chat-latest" || (providerID === "openrouter" && modelID === "openai/gpt-5-chat"))
00801| delete provider.models[modelID]
00802| if (model.status === "alpha" && !Flag.OPENCODE_ENABLE_EXPERIMENTAL_MODELS) delete provider.models[modelID]
00803| if (
00804| (configProvider?.blacklist && configProvider.blacklist.includes(modelID)) ||
00805| (configProvider?.whitelist && !configProvider.whitelist.includes(modelID))
00806| )
00807| delete provider.models[modelID]
00808| }
00809|
00810| if (Object.keys(provider.models).length === 0) {
00811| delete providers[providerID]
00812| continue
00813| }
00814|
00815| log.info("found", { providerID })
00816| }
00817|
00818| return {
00819| models: languages,
00820| providers,
00821| sdk,
00822| modelLoaders,
00823| }
00824| })
00825|
(File has more lines. Use 'offset' parameter to read beyond line 825)
</file>
---
## Assistant
Now let me check how the models database is structured and where the npm packages come from:
Tool: grep
Input:
{
"pattern": "cloudflare",
"path": "/Users/englishm/git/opencode/packages/opencode/src",
"include": "*.json"
}Output:
No files found
---
## Assistant
Tool: glob
Input:
{
"pattern": "**/models*.json",
"path": "/Users/englishm/git/opencode/packages/opencode"
}Output:
No files found
---
## Assistant
Tool: grep
Input:
{
"pattern": "database\\[",
"path": "/Users/englishm/git/opencode/packages/opencode/src/provider"
}Output:
Found 9 matches
/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts:
Line 590: if (database["github-copilot"]) {
Line 591: const githubCopilot = database["github-copilot"]
Line 592: database["github-copilot-enterprise"] = {
Line 610: const match = database[providerID]
Line 618: const existing = database[providerID]
Line 686: database[providerID] = parsed
Line 733: const options = await plugin.auth.loader(() => Auth.get(providerID) as any, database[plugin.auth.provider])
Line 748: database[enterpriseProviderID],
Line 761: const result = await fn(database[providerID])
---
## Assistant
Tool: grep
Input:
{
"pattern": "type Database|interface Database|Database =\"",
"path": "/Users/englishm/git/opencode/packages/opencode/src/provider"
}Output:
No files found
---
## Assistant
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 550,
"limit": 50
}Output:
<file>
00551|
00552| export function fromModelsDevProvider(provider: ModelsDev.Provider): Info {
00553| return {
00554| id: provider.id,
00555| source: "custom",
00556| name: provider.name,
00557| env: provider.env ?? [],
00558| options: {},
00559| models: mapValues(provider.models, (model) => fromModelsDevModel(provider, model)),
00560| }
00561| }
00562|
00563| const state = Instance.state(async () => {
00564| using _ = log.time("state")
00565| const config = await Config.get()
00566| const modelsDev = await ModelsDev.get()
00567| const database = mapValues(modelsDev, fromModelsDevProvider)
00568|
00569| const disabled = new Set(config.disabled_providers ?? [])
00570| const enabled = config.enabled_providers ? new Set(config.enabled_providers) : null
00571|
00572| function isProviderAllowed(providerID: string): boolean {
00573| if (enabled && !enabled.has(providerID)) return false
00574| if (disabled.has(providerID)) return false
00575| return true
00576| }
00577|
00578| const providers: { [providerID: string]: Info } = {}
00579| const languages = new Map<string, LanguageModelV2>()
00580| const modelLoaders: {
00581| [providerID: string]: CustomModelLoader
00582| } = {}
00583| const sdk = new Map<number, SDK>()
00584|
00585| log.info("init")
00586|
00587| const configProviders = Object.entries(config.provider ?? {})
00588|
00589| // Add GitHub Copilot Enterprise provider that inherits from GitHub Copilot
00590| if (database["github-copilot"]) {
00591| const githubCopilot = database["github-copilot"]
00592| database["github-copilot-enterprise"] = {
00593| ...githubCopilot,
00594| id: "github-copilot-enterprise",
00595| name: "GitHub Copilot Enterprise",
00596| models: mapValues(githubCopilot.models, (model) => ({
00597| ...model,
00598| providerID: "github-copilot-enterprise",
00599| })),
00600| }
(File has more lines. Use 'offset' parameter to read beyond line 600)
</file>
---
## Assistant
Let me look at how ModelsDev works and what defines the npm package:
Tool: grep
Input:
{
"pattern": "ModelsDev",
"path": "/Users/englishm/git/opencode/packages/opencode/src"
}Output:
Found 26 matches
/Users/englishm/git/opencode/packages/opencode/src/server/server.ts:
Line 15: import { ModelsDev } from "../provider/models"
Line 1623: all: ModelsDev.Provider.array(),
Line 1638: const allProviders = await ModelsDev.get()
Line 1648: mapValues(filteredProviders, (x) => Provider.fromModelsDevProvider(x)),
/Users/englishm/git/opencode/packages/opencode/src/provider/transform.ts:
Line 5: import type { ModelsDev } from "./models"
Line 7: type Modality = NonNullable<ModelsDev.Model["modalities"]>["input"][number]
/Users/englishm/git/opencode/packages/opencode/src/provider/models.ts:
Line 9: export namespace ModelsDev {
Line 106: setInterval(() => ModelsDev.refresh(), 60 * 1000 * 60).unref()
/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts:
Line 9: import { ModelsDev } from "./models"
Line 491: function fromModelsDevModel(provider: ModelsDev.Provider, model: ModelsDev.Model): Model {
Line 552: export function fromModelsDevProvider(provider: ModelsDev.Provider): Info {
Line 559: models: mapValues(provider.models, (model) => fromModelsDevModel(provider, model)),
Line 566: const modelsDev = await ModelsDev.get()
Line 567: const database = mapValues(modelsDev, fromModelsDevProvider)
/Users/englishm/git/opencode/packages/opencode/src/config/config.ts:
Line 7: import { ModelsDev } from "../provider/models"
Line 618: export const Provider = ModelsDev.Provider.partial()
Line 622: models: z.record(z.string(), ModelsDev.Model.partial()).optional(),
/Users/englishm/git/opencode/packages/opencode/src/cli/cmd/github.ts:
Line 20: import { ModelsDev } from "../../provider/models"
Line 204: const providers = await ModelsDev.get().then((p) => {
/Users/englishm/git/opencode/packages/opencode/src/cli/cmd/models.ts:
Line 4: import { ModelsDev } from "../../provider/models"
Line 30: await ModelsDev.refresh()
/Users/englishm/git/opencode/packages/opencode/src/cli/cmd/auth.ts:
Line 5: import { ModelsDev } from "../../provider/models"
Line 181: const database = await ModelsDev.get()
Line 254: await ModelsDev.refresh().catch(() => {})
Line 261: const providers = await ModelsDev.get().then((x) => {
Line 379: const database = await ModelsDev.get()
---
## Assistant
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/models.ts"
}Output:
<file>
00001| import { Global } from "../global"
00002| import { Log } from "../util/log"
00003| import path from "path"
00004| import z from "zod"
00005| import { data } from "./models-macro" with { type: "macro" }
00006| import { Installation } from "../installation"
00007| import { Flag } from "../flag/flag"
00008|
00009| export namespace ModelsDev {
00010| const log = Log.create({ service: "models.dev" })
00011| const filepath = path.join(Global.Path.cache, "models.json")
00012|
00013| export const Model = z.object({
00014| id: z.string(),
00015| name: z.string(),
00016| family: z.string().optional(),
00017| release_date: z.string(),
00018| attachment: z.boolean(),
00019| reasoning: z.boolean(),
00020| temperature: z.boolean(),
00021| tool_call: z.boolean(),
00022| interleaved: z
00023| .union([
00024| z.literal(true),
00025| z
00026| .object({
00027| field: z.enum(["reasoning_content", "reasoning_details"]),
00028| })
00029| .strict(),
00030| ])
00031| .optional(),
00032| cost: z
00033| .object({
00034| input: z.number(),
00035| output: z.number(),
00036| cache_read: z.number().optional(),
00037| cache_write: z.number().optional(),
00038| context_over_200k: z
00039| .object({
00040| input: z.number(),
00041| output: z.number(),
00042| cache_read: z.number().optional(),
00043| cache_write: z.number().optional(),
00044| })
00045| .optional(),
00046| })
00047| .optional(),
00048| limit: z.object({
00049| context: z.number(),
00050| output: z.number(),
00051| }),
00052| modalities: z
00053| .object({
00054| input: z.array(z.enum(["text", "audio", "image", "video", "pdf"])),
00055| output: z.array(z.enum(["text", "audio", "image", "video", "pdf"])),
00056| })
00057| .optional(),
00058| experimental: z.boolean().optional(),
00059| status: z.enum(["alpha", "beta", "deprecated"]).optional(),
00060| options: z.record(z.string(), z.any()),
00061| headers: z.record(z.string(), z.string()).optional(),
00062| provider: z.object({ npm: z.string() }).optional(),
00063| })
00064| export type Model = z.infer<typeof Model>
00065|
00066| export const Provider = z.object({
00067| api: z.string().optional(),
00068| name: z.string(),
00069| env: z.array(z.string()),
00070| id: z.string(),
00071| npm: z.string().optional(),
00072| models: z.record(z.string(), Model),
00073| })
00074|
00075| export type Provider = z.infer<typeof Provider>
00076|
00077| export async function get() {
00078| refresh()
00079| const file = Bun.file(filepath)
00080| const result = await file.json().catch(() => {})
00081| if (result) return result as Record<string, Provider>
00082| const json = await data()
00083| return JSON.parse(json) as Record<string, Provider>
00084| }
00085|
00086| export async function refresh() {
00087| if (Flag.OPENCODE_DISABLE_MODELS_FETCH) return
00088| const file = Bun.file(filepath)
00089| log.info("refreshing", {
00090| file,
00091| })
00092| const result = await fetch("https://models.dev/api.json", {
00093| headers: {
00094| "User-Agent": Installation.USER_AGENT,
00095| },
00096| signal: AbortSignal.timeout(10 * 1000),
00097| }).catch((e) => {
00098| log.error("Failed to fetch models.dev", {
00099| error: e,
00100| })
00101| })
00102| if (result && result.ok) await Bun.write(file, await result.text())
00103| }
00104| }
00105|
00106| setInterval(() => ModelsDev.refresh(), 60 * 1000 * 60).unref()
00107|
(End of file - total 107 lines)
</file>
---
## Assistant
Now let me see how `fromModelsDevModel` handles the npm package:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 485,
"limit": 70
}Output:
<file>
00486| .meta({
00487| ref: "Provider",
00488| })
00489| export type Info = z.infer<typeof Info>
00490|
00491| function fromModelsDevModel(provider: ModelsDev.Provider, model: ModelsDev.Model): Model {
00492| return {
00493| id: model.id,
00494| providerID: provider.id,
00495| name: model.name,
00496| family: model.family,
00497| api: {
00498| id: model.id,
00499| url: provider.api!,
00500| npm: model.provider?.npm ?? provider.npm ?? provider.id,
00501| },
00502| status: model.status ?? "active",
00503| headers: model.headers ?? {},
00504| options: model.options ?? {},
00505| cost: {
00506| input: model.cost?.input ?? 0,
00507| output: model.cost?.output ?? 0,
00508| cache: {
00509| read: model.cost?.cache_read ?? 0,
00510| write: model.cost?.cache_write ?? 0,
00511| },
00512| experimentalOver200K: model.cost?.context_over_200k
00513| ? {
00514| cache: {
00515| read: model.cost.context_over_200k.cache_read ?? 0,
00516| write: model.cost.context_over_200k.cache_write ?? 0,
00517| },
00518| input: model.cost.context_over_200k.input,
00519| output: model.cost.context_over_200k.output,
00520| }
00521| : undefined,
00522| },
00523| limit: {
00524| context: model.limit.context,
00525| output: model.limit.output,
00526| },
00527| capabilities: {
00528| temperature: model.temperature,
00529| reasoning: model.reasoning,
00530| attachment: model.attachment,
00531| toolcall: model.tool_call,
00532| input: {
00533| text: model.modalities?.input?.includes("text") ?? false,
00534| audio: model.modalities?.input?.includes("audio") ?? false,
00535| image: model.modalities?.input?.includes("image") ?? false,
00536| video: model.modalities?.input?.includes("video") ?? false,
00537| pdf: model.modalities?.input?.includes("pdf") ?? false,
00538| },
00539| output: {
00540| text: model.modalities?.output?.includes("text") ?? false,
00541| audio: model.modalities?.output?.includes("audio") ?? false,
00542| image: model.modalities?.output?.includes("image") ?? false,
00543| video: model.modalities?.output?.includes("video") ?? false,
00544| pdf: model.modalities?.output?.includes("pdf") ?? false,
00545| },
00546| interleaved: model.interleaved ?? false,
00547| },
00548| release_date: model.release_date,
00549| }
00550| }
00551|
00552| export function fromModelsDevProvider(provider: ModelsDev.Provider): Info {
00553| return {
00554| id: provider.id,
00555| source: "custom",
(File has more lines. Use 'offset' parameter to read beyond line 555)
</file>
---
## Assistant
Now I understand the issue better. Let me check how the SDK is loaded for providers and how the `cloudflare-ai-gateway` provider is handled:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 830,
"limit": 100
}Output:
<file>
00831| try {
00832| using _ = log.time("getSDK", {
00833| providerID: model.providerID,
00834| })
00835| const s = await state()
00836| const provider = s.providers[model.providerID]
00837| const options = { ...provider.options }
00838|
00839| if (model.api.npm.includes("@ai-sdk/openai-compatible") && options["includeUsage"] !== false) {
00840| options["includeUsage"] = true
00841| }
00842|
00843| if (!options["baseURL"]) options["baseURL"] = model.api.url
00844| if (options["apiKey"] === undefined && provider.key) options["apiKey"] = provider.key
00845| if (model.headers)
00846| options["headers"] = {
00847| ...options["headers"],
00848| ...model.headers,
00849| }
00850|
00851| const key = Bun.hash.xxHash32(JSON.stringify({ npm: model.api.npm, options }))
00852| const existing = s.sdk.get(key)
00853| if (existing) return existing
00854|
00855| const customFetch = options["fetch"]
00856|
00857| options["fetch"] = async (input: any, init?: BunFetchRequestInit) => {
00858| // Preserve custom fetch if it exists, wrap it with timeout logic
00859| const fetchFn = customFetch ?? fetch
00860| const opts = init ?? {}
00861|
00862| if (options["timeout"] !== undefined && options["timeout"] !== null) {
00863| const signals: AbortSignal[] = []
00864| if (opts.signal) signals.push(opts.signal)
00865| if (options["timeout"] !== false) signals.push(AbortSignal.timeout(options["timeout"]))
00866|
00867| const combined = signals.length > 1 ? AbortSignal.any(signals) : signals[0]
00868|
00869| opts.signal = combined
00870| }
00871|
00872| return fetchFn(input, {
00873| ...opts,
00874| // @ts-ignore see here: https://github.com/oven-sh/bun/issues/16682
00875| timeout: false,
00876| })
00877| }
00878|
00879| // Special case: google-vertex-anthropic uses a subpath import
00880| const bundledKey =
00881| model.providerID === "google-vertex-anthropic" ? "@ai-sdk/google-vertex/anthropic" : model.api.npm
00882| const bundledFn = BUNDLED_PROVIDERS[bundledKey]
00883| if (bundledFn) {
00884| log.info("using bundled provider", { providerID: model.providerID, pkg: bundledKey })
00885| const loaded = bundledFn({
00886| name: model.providerID,
00887| ...options,
00888| })
00889| s.sdk.set(key, loaded)
00890| return loaded as SDK
00891| }
00892|
00893| let installedPath: string
00894| if (!model.api.npm.startsWith("file://")) {
00895| installedPath = await BunProc.install(model.api.npm, "latest")
00896| } else {
00897| log.info("loading local provider", { pkg: model.api.npm })
00898| installedPath = model.api.npm
00899| }
00900|
00901| const mod = await import(installedPath)
00902|
00903| const fn = mod[Object.keys(mod).find((key) => key.startsWith("create"))!]
00904| const loaded = fn({
00905| name: model.providerID,
00906| ...options,
00907| })
00908| s.sdk.set(key, loaded)
00909| return loaded as SDK
00910| } catch (e) {
00911| throw new InitError({ providerID: model.providerID }, { cause: e })
00912| }
00913| }
00914|
00915| export async function getProvider(providerID: string) {
00916| return state().then((s) => s.providers[providerID])
00917| }
00918|
00919| export async function getModel(providerID: string, modelID: string) {
00920| const s = await state()
00921| const provider = s.providers[providerID]
00922| if (!provider) {
00923| const availableProviders = Object.keys(s.providers)
00924| const matches = fuzzysort.go(providerID, availableProviders, { limit: 3, threshold: -10000 })
00925| const suggestions = matches.map((m) => m.target)
00926| throw new ModelNotFoundError({ providerID, modelID, suggestions })
00927| }
00928|
00929| const info = provider.models[modelID]
00930| if (!info) {
(File has more lines. Use 'offset' parameter to read beyond line 930)
</file>
---
## Assistant
Now I see the flow. Let me check how models get loaded with the custom model loader:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 940,
"limit": 60
}Output:
<file>
00941| const key = `${model.providerID}/${model.id}`
00942| if (s.models.has(key)) return s.models.get(key)!
00943|
00944| const provider = s.providers[model.providerID]
00945| const sdk = await getSDK(model)
00946|
00947| try {
00948| const language = s.modelLoaders[model.providerID]
00949| ? await s.modelLoaders[model.providerID](sdk, model.api.id, provider.options)
00950| : sdk.languageModel(model.api.id)
00951| s.models.set(key, language)
00952| return language
00953| } catch (e) {
00954| if (e instanceof NoSuchModelError)
00955| throw new ModelNotFoundError(
00956| {
00957| modelID: model.id,
00958| providerID: model.providerID,
00959| },
00960| { cause: e },
00961| )
00962| throw e
00963| }
00964| }
00965|
00966| export async function closest(providerID: string, query: string[]) {
00967| const s = await state()
00968| const provider = s.providers[providerID]
00969| if (!provider) return undefined
00970| for (const item of query) {
00971| for (const modelID of Object.keys(provider.models)) {
00972| if (modelID.includes(item))
00973| return {
00974| providerID,
00975| modelID,
00976| }
00977| }
00978| }
00979| }
00980|
00981| export async function getSmallModel(providerID: string) {
00982| const cfg = await Config.get()
00983|
00984| if (cfg.small_model) {
00985| const parsed = parseModel(cfg.small_model)
00986| return getModel(parsed.providerID, parsed.modelID)
00987| }
00988|
00989| const provider = await state().then((state) => state.providers[providerID])
00990| if (provider) {
00991| let priority = [
00992| "claude-haiku-4-5",
00993| "claude-haiku-4.5",
00994| "3-5-haiku",
00995| "3.5-haiku",
00996| "gemini-2.5-flash",
00997| "gpt-5-nano",
00998| ]
00999| // claude-haiku-4.5 is considered a premium model in github copilot, we shouldn't use premium requests for title gen
01000| if (providerID === "github-copilot") {
(File has more lines. Use 'offset' parameter to read beyond line 1000)
</file>
---
## Assistant
Now I see the issue! The `modelLoaders` is populated in the provider state, and the custom `getModel` function from the `cloudflare-ai-gateway` loader calls `sdk.chat(modelID)`. But the problem is the SDK being loaded doesn't have a `.chat()` method because the wrong npm package is being used.
Let me check where `modelLoaders` gets populated:
Tool: grep
Input:
{
"pattern": "modelLoaders\\[",
"path": "/Users/englishm/git/opencode/packages/opencode/src/provider"
}Output:
Found 3 matches
/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts:
Line 763: if (result.getModel) modelLoaders[providerID] = result.getModel
Line 948: const language = s.modelLoaders[model.providerID]
Line 949: ? await s.modelLoaders[model.providerID](sdk, model.api.id, provider.options)
---
## Assistant
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 755,
"limit": 30
}Output:
<file>
00756| }
00757| }
00758|
00759| for (const [providerID, fn] of Object.entries(CUSTOM_LOADERS)) {
00760| if (disabled.has(providerID)) continue
00761| const result = await fn(database[providerID])
00762| if (result && (result.autoload || providers[providerID])) {
00763| if (result.getModel) modelLoaders[providerID] = result.getModel
00764| mergeProvider(providerID, {
00765| source: "custom",
00766| options: result.options,
00767| })
00768| }
00769| }
00770|
00771| // load config
00772| for (const [providerID, provider] of configProviders) {
00773| const partial: Partial<Info> = { source: "config" }
00774| if (provider.env) partial.env = provider.env
00775| if (provider.name) partial.name = provider.name
00776| if (provider.options) partial.options = provider.options
00777| mergeProvider(providerID, partial)
00778| }
00779|
00780| for (const [providerID, provider] of Object.entries(providers)) {
00781| if (!isProviderAllowed(providerID)) {
00782| delete providers[providerID]
00783| continue
00784| }
00785|
(File has more lines. Use 'offset' parameter to read beyond line 785)
</file>
---
## Assistant
Now I understand the full picture. The issue is:
1. `cloudflare-ai-gateway` is **NOT in the models.dev database** (it's a gateway, not a direct provider)
2. The custom loader at line 356-394 has a `getModel` function that calls `sdk.chat(modelID)`
3. But when the models are configured (via config file), they don't have a proper `npm` package set
4. So the fallback at line 500 uses `provider.npm ?? provider.id` which would be `"cloudflare-ai-gateway"` - a package that doesn't exist!
Let me verify by checking what models.dev returns for cloudflare:
Tool: bash
Input:
{
"command": "curl -s \"https://models.dev/api.json\" | jq 'keys | map(select(test(\"cloudflare\"; \"i\")))'",
"description": "Check if cloudflare exists in models.dev"
}Output:
[
"cloudflare-ai-gateway",
"cloudflare-workers-ai"
]
---
## Assistant
Tool: bash
Input:
{
"command": "curl -s \"https://models.dev/api.json\" | jq '.[\"cloudflare-ai-gateway\"]'",
"description": "Check cloudflare-ai-gateway provider details"
}Output:
{
"id": "cloudflare-ai-gateway",
"env": [
"CLOUDFLARE_API_TOKEN",
"CLOUDFLARE_ACCOUNT_ID",
"CLOUDFLARE_GATEWAY_ID"
],
"npm": "@ai-sdk/openai-compatible",
"api": "https://gateway.ai.cloudflare.com/v1/${CLOUDFLARE_ACCOUNT_ID}/${CLOUDFLARE_GATEWAY_ID}/compat/",
"name": "Cloudflare AI Gateway",
"doc": "https://developers.cloudflare.com/ai-gateway/",
"models": {
"workers-ai/@cf/ibm-granite/granite-4.0-h-micro": {
"id": "workers-ai/@cf/ibm-granite/granite-4.0-h-micro",
"name": "IBM Granite 4.0 H Micro",
"family": "granite-4",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-10-15",
"last_updated": "2025-10-15",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.017,
"output": 0.11
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/facebook/bart-large-cnn": {
"id": "workers-ai/@cf/facebook/bart-large-cnn",
"name": "BART Large CNN",
"family": "bart",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-09",
"last_updated": "2025-04-09",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/mistral/mistral-7b-instruct-v0.1": {
"id": "workers-ai/@cf/mistral/mistral-7b-instruct-v0.1",
"name": "Mistral 7B Instruct v0.1",
"family": "mistral-7b",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.11,
"output": 0.19
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/huggingface/distilbert-sst-2-int8": {
"id": "workers-ai/@cf/huggingface/distilbert-sst-2-int8",
"name": "DistilBERT SST-2 INT8",
"family": "distilbert",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.026,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/myshell-ai/melotts": {
"id": "workers-ai/@cf/myshell-ai/melotts",
"name": "MyShell MeloTTS",
"family": "melotts",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-11-14",
"last_updated": "2025-11-14",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/google/gemma-3-12b-it": {
"id": "workers-ai/@cf/google/gemma-3-12b-it",
"name": "Gemma 3 12B IT",
"family": "gemma-3",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-11",
"last_updated": "2025-04-11",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.35,
"output": 0.56
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/pfnet/plamo-embedding-1b": {
"id": "workers-ai/@cf/pfnet/plamo-embedding-1b",
"name": "PLaMo Embedding 1B",
"family": "plamo-embedding",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-09-25",
"last_updated": "2025-09-25",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.019,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/openai/gpt-oss-20b": {
"id": "workers-ai/@cf/openai/gpt-oss-20b",
"name": "GPT OSS 20B",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-08-05",
"last_updated": "2025-08-05",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.2,
"output": 0.3
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/openai/gpt-oss-120b": {
"id": "workers-ai/@cf/openai/gpt-oss-120b",
"name": "GPT OSS 120B",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-08-05",
"last_updated": "2025-08-05",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.35,
"output": 0.75
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/ai4bharat/indictrans2-en-indic-1B": {
"id": "workers-ai/@cf/ai4bharat/indictrans2-en-indic-1B",
"name": "IndicTrans2 EN-Indic 1B",
"family": "indictrans2",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-09-25",
"last_updated": "2025-09-25",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.34,
"output": 0.34
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/pipecat-ai/smart-turn-v2": {
"id": "workers-ai/@cf/pipecat-ai/smart-turn-v2",
"name": "Pipecat Smart Turn v2",
"family": "smart-turn",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-11-14",
"last_updated": "2025-11-14",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/qwen/qwen2.5-coder-32b-instruct": {
"id": "workers-ai/@cf/qwen/qwen2.5-coder-32b-instruct",
"name": "Qwen 2.5 Coder 32B Instruct",
"family": "qwen2.5-coder",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-11",
"last_updated": "2025-04-11",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.66,
"output": 1
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/qwen/qwen3-30b-a3b-fp8": {
"id": "workers-ai/@cf/qwen/qwen3-30b-a3b-fp8",
"name": "Qwen3 30B A3B FP8",
"family": "qwen3",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-11-14",
"last_updated": "2025-11-14",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.051,
"output": 0.34
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/qwen/qwen3-embedding-0.6b": {
"id": "workers-ai/@cf/qwen/qwen3-embedding-0.6b",
"name": "Qwen3 Embedding 0.6B",
"family": "qwen3-embedding",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-11-14",
"last_updated": "2025-11-14",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.012,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/qwen/qwq-32b": {
"id": "workers-ai/@cf/qwen/qwq-32b",
"name": "QwQ 32B",
"family": "qwq",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-11",
"last_updated": "2025-04-11",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.66,
"output": 1
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/mistralai/mistral-small-3.1-24b-instruct": {
"id": "workers-ai/@cf/mistralai/mistral-small-3.1-24b-instruct",
"name": "Mistral Small 3.1 24B Instruct",
"family": "mistral-small",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-11",
"last_updated": "2025-04-11",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.35,
"output": 0.56
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/deepgram/aura-2-es": {
"id": "workers-ai/@cf/deepgram/aura-2-es",
"name": "Deepgram Aura 2 (ES)",
"family": "aura-2",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-11-14",
"last_updated": "2025-11-14",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/deepgram/aura-2-en": {
"id": "workers-ai/@cf/deepgram/aura-2-en",
"name": "Deepgram Aura 2 (EN)",
"family": "aura-2",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-11-14",
"last_updated": "2025-11-14",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/deepgram/nova-3": {
"id": "workers-ai/@cf/deepgram/nova-3",
"name": "Deepgram Nova 3",
"family": "nova",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-11-14",
"last_updated": "2025-11-14",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/aisingapore/gemma-sea-lion-v4-27b-it": {
"id": "workers-ai/@cf/aisingapore/gemma-sea-lion-v4-27b-it",
"name": "Gemma SEA-LION v4 27B IT",
"family": "gemma-sea-lion",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-09-25",
"last_updated": "2025-09-25",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.35,
"output": 0.56
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3.2-11b-vision-instruct": {
"id": "workers-ai/@cf/meta/llama-3.2-11b-vision-instruct",
"name": "Llama 3.2 11B Vision Instruct",
"family": "llama-3.2-vision",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.049,
"output": 0.68
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3.1-8b-instruct-fp8": {
"id": "workers-ai/@cf/meta/llama-3.1-8b-instruct-fp8",
"name": "Llama 3.1 8B Instruct FP8",
"family": "llama-3.1",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.15,
"output": 0.29
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-2-7b-chat-fp16": {
"id": "workers-ai/@cf/meta/llama-2-7b-chat-fp16",
"name": "Llama 2 7B Chat FP16",
"family": "llama-2",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.56,
"output": 6.67
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3-8b-instruct": {
"id": "workers-ai/@cf/meta/llama-3-8b-instruct",
"name": "Llama 3 8B Instruct",
"family": "llama-3",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.28,
"output": 0.83
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3.1-8b-instruct": {
"id": "workers-ai/@cf/meta/llama-3.1-8b-instruct",
"name": "Llama 3.1 8B Instruct",
"family": "llama-3.1",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.28,
"output": 0.8299999999999998
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/m2m100-1.2b": {
"id": "workers-ai/@cf/meta/m2m100-1.2b",
"name": "M2M100 1.2B",
"family": "m2m100",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.34,
"output": 0.34
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3.2-3b-instruct": {
"id": "workers-ai/@cf/meta/llama-3.2-3b-instruct",
"name": "Llama 3.2 3B Instruct",
"family": "llama-3.2",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.051,
"output": 0.34
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3.3-70b-instruct-fp8-fast": {
"id": "workers-ai/@cf/meta/llama-3.3-70b-instruct-fp8-fast",
"name": "Llama 3.3 70B Instruct FP8 Fast",
"family": "llama-3.3",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.29,
"output": 2.25
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3-8b-instruct-awq": {
"id": "workers-ai/@cf/meta/llama-3-8b-instruct-awq",
"name": "Llama 3 8B Instruct AWQ",
"family": "llama-3",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.12,
"output": 0.27
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3.2-1b-instruct": {
"id": "workers-ai/@cf/meta/llama-3.2-1b-instruct",
"name": "Llama 3.2 1B Instruct",
"family": "llama-3.2",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.027,
"output": 0.2
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-4-scout-17b-16e-instruct": {
"id": "workers-ai/@cf/meta/llama-4-scout-17b-16e-instruct",
"name": "Llama 4 Scout 17B 16E Instruct",
"family": "llama-4-scout",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-16",
"last_updated": "2025-04-16",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.27,
"output": 0.85
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-guard-3-8b": {
"id": "workers-ai/@cf/meta/llama-guard-3-8b",
"name": "Llama Guard 3 8B",
"family": "llama-guard",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.48,
"output": 0.03
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/meta/llama-3.1-8b-instruct-awq": {
"id": "workers-ai/@cf/meta/llama-3.1-8b-instruct-awq",
"name": "Llama 3.1 8B Instruct AWQ",
"family": "llama-3.1",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.12,
"output": 0.27
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/baai/bge-m3": {
"id": "workers-ai/@cf/baai/bge-m3",
"name": "BGE M3",
"family": "bge-m3",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.012,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/baai/bge-base-en-v1.5": {
"id": "workers-ai/@cf/baai/bge-base-en-v1.5",
"name": "BGE Base EN v1.5",
"family": "bge-base",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.067,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/baai/bge-large-en-v1.5": {
"id": "workers-ai/@cf/baai/bge-large-en-v1.5",
"name": "BGE Large EN v1.5",
"family": "bge-large",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.2,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/baai/bge-reranker-base": {
"id": "workers-ai/@cf/baai/bge-reranker-base",
"name": "BGE Reranker Base",
"family": "bge-reranker",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-09",
"last_updated": "2025-04-09",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.0031,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/baai/bge-small-en-v1.5": {
"id": "workers-ai/@cf/baai/bge-small-en-v1.5",
"name": "BGE Small EN v1.5",
"family": "bge-small",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.02,
"output": 0
},
"limit": {
"context": 128000,
"output": 16384
}
},
"workers-ai/@cf/deepseek-ai/deepseek-r1-distill-qwen-32b": {
"id": "workers-ai/@cf/deepseek-ai/deepseek-r1-distill-qwen-32b",
"name": "DeepSeek R1 Distill Qwen 32B",
"family": "deepseek-r1-distill-qwen",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": true,
"release_date": "2025-04-03",
"last_updated": "2025-04-03",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.5,
"output": 4.88
},
"limit": {
"context": 128000,
"output": 16384
}
},
"openai/gpt-4": {
"id": "openai/gpt-4",
"name": "GPT-4",
"family": "gpt-4",
"attachment": true,
"reasoning": false,
"tool_call": true,
"structured_output": false,
"temperature": true,
"knowledge": "2023-11",
"release_date": "2023-11-06",
"last_updated": "2024-04-09",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 30,
"output": 60
},
"limit": {
"context": 8192,
"output": 8192
}
},
"openai/gpt-5.1-codex": {
"id": "openai/gpt-5.1-codex",
"name": "GPT-5.1 Codex",
"family": "gpt-5-codex",
"attachment": true,
"reasoning": true,
"tool_call": true,
"structured_output": true,
"temperature": false,
"knowledge": "2024-09-30",
"release_date": "2025-11-13",
"last_updated": "2025-11-13",
"modalities": {
"input": [
"text",
"image"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 1.25,
"output": 10,
"cache_read": 0.125
},
"limit": {
"context": 400000,
"output": 128000
}
},
"openai/gpt-3.5-turbo": {
"id": "openai/gpt-3.5-turbo",
"name": "GPT-3.5-turbo",
"family": "gpt-3.5-turbo",
"attachment": false,
"reasoning": false,
"tool_call": false,
"structured_output": false,
"temperature": true,
"knowledge": "2021-09-01",
"release_date": "2023-03-01",
"last_updated": "2023-11-06",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 0.5,
"output": 1.5,
"cache_read": 1.25
},
"limit": {
"context": 16385,
"output": 4096
}
},
"openai/gpt-4-turbo": {
"id": "openai/gpt-4-turbo",
"name": "GPT-4 Turbo",
"family": "gpt-4-turbo",
"attachment": true,
"reasoning": false,
"tool_call": true,
"structured_output": false,
"temperature": true,
"knowledge": "2023-12",
"release_date": "2023-11-06",
"last_updated": "2024-04-09",
"modalities": {
"input": [
"text",
"image"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 10,
"output": 30
},
"limit": {
"context": 128000,
"output": 4096
}
},
"openai/o3-mini": {
"id": "openai/o3-mini",
"name": "o3-mini",
"family": "o3-mini",
"attachment": false,
"reasoning": true,
"tool_call": true,
"structured_output": true,
"temperature": false,
"knowledge": "2024-05",
"release_date": "2024-12-20",
"last_updated": "2025-01-29",
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 1.1,
"output": 4.4,
"cache_read": 0.55
},
"limit": {
"context": 200000,
"output": 100000
}
},
"openai/gpt-5.1": {
"id": "openai/gpt-5.1",
"name": "GPT-5.1",
"family": "gpt-5",
"attachment": true,
"reasoning": true,
"tool_call": true,
"temperature": false,
"knowledge": "2024-09-30",
"release_date": "2025-11-13",
"last_updated": "2025-11-13",
"modalities": {
"input": [
"text",
"image"
],
"output": [
"text"
]
},
"open_weights": false,
"cost": {
"input": 1.25,
"output": 10
<bash_metadata>
bash tool truncated output as it exceeded 30000 char limit
</bash_metadata>
---
## Assistant
The models.dev database **does** include `cloudflare-ai-gateway` and it specifies `"npm": "@ai-sdk/openai-compatible"`. But the issue is likely that the custom `getModel` function at line 373-374 calls `sdk.chat(modelID)`, but the `@ai-sdk/openai-compatible` package might not have a `.chat()` method in the same way - it likely uses `.languageModel()` instead.
Let me check what the `@ai-sdk/openai-compatible` package API looks like:
Tool: bash
Input:
{
"command": "curl -s \"https://unpkg.com/@ai-sdk/openai-compatible@latest/dist/index.d.ts\" | head -100",
"description": "Check openai-compatible SDK API"
}Output:
Redirecting to /@ai-sdk/openai-compatible@2.0.1/dist/index.d.ts
---
## Assistant
Tool: bash
Input:
{
"command": "curl -sL \"https://unpkg.com/@ai-sdk/openai-compatible@latest/dist/index.d.ts\" | head -150",
"description": "Check openai-compatible SDK API"
}Output:
import { SharedV3ProviderMetadata, LanguageModelV3, LanguageModelV3CallOptions, LanguageModelV3GenerateResult, LanguageModelV3StreamResult, EmbeddingModelV3, ImageModelV3, ProviderV3 } from '@ai-sdk/provider';
import { FetchFunction } from '@ai-sdk/provider-utils';
import { ZodType, z } from 'zod/v4';
declare const openaiCompatibleErrorDataSchema: z.ZodObject<{
error: z.ZodObject<{
message: z.ZodString;
type: z.ZodOptional<z.ZodNullable<z.ZodString>>;
param: z.ZodOptional<z.ZodNullable<z.ZodAny>>;
code: z.ZodOptional<z.ZodNullable<z.ZodUnion<readonly [z.ZodString, z.ZodNumber]>>>;
}, z.core.$strip>;
}, z.core.$strip>;
type OpenAICompatibleErrorData = z.infer<typeof openaiCompatibleErrorDataSchema>;
type ProviderErrorStructure<T> = {
errorSchema: ZodType<T>;
errorToMessage: (error: T) => string;
isRetryable?: (response: Response, error?: T) => boolean;
};
type OpenAICompatibleChatModelId = string;
declare const openaiCompatibleProviderOptions: z.ZodObject<{
user: z.ZodOptional<z.ZodString>;
reasoningEffort: z.ZodOptional<z.ZodString>;
textVerbosity: z.ZodOptional<z.ZodString>;
}, z.core.$strip>;
type OpenAICompatibleProviderOptions = z.infer<typeof openaiCompatibleProviderOptions>;
/**
Extracts provider-specific metadata from API responses.
Used to standardize metadata handling across different LLM providers while allowing
provider-specific metadata to be captured.
*/
type MetadataExtractor = {
/**
* Extracts provider metadata from a complete, non-streaming response.
*
* @param parsedBody - The parsed response JSON body from the provider's API.
*
* @returns Provider-specific metadata or undefined if no metadata is available.
* The metadata should be under a key indicating the provider id.
*/
extractMetadata: ({ parsedBody, }: {
parsedBody: unknown;
}) => Promise<SharedV3ProviderMetadata | undefined>;
/**
* Creates an extractor for handling streaming responses. The returned object provides
* methods to process individual chunks and build the final metadata from the accumulated
* stream data.
*
* @returns An object with methods to process chunks and build metadata from a stream
*/
createStreamExtractor: () => {
/**
* Process an individual chunk from the stream. Called for each chunk in the response stream
* to accumulate metadata throughout the streaming process.
*
* @param parsedChunk - The parsed JSON response chunk from the provider's API
*/
processChunk(parsedChunk: unknown): void;
/**
* Builds the metadata object after all chunks have been processed.
* Called at the end of the stream to generate the complete provider metadata.
*
* @returns Provider-specific metadata or undefined if no metadata is available.
* The metadata should be under a key indicating the provider id.
*/
buildMetadata(): SharedV3ProviderMetadata | undefined;
};
};
type OpenAICompatibleChatConfig = {
provider: string;
headers: () => Record<string, string | undefined>;
url: (options: {
modelId: string;
path: string;
}) => string;
fetch?: FetchFunction;
includeUsage?: boolean;
errorStructure?: ProviderErrorStructure<any>;
metadataExtractor?: MetadataExtractor;
/**
* Whether the model supports structured outputs.
*/
supportsStructuredOutputs?: boolean;
/**
* The supported URLs for the model.
*/
supportedUrls?: () => LanguageModelV3['supportedUrls'];
};
declare class OpenAICompatibleChatLanguageModel implements LanguageModelV3 {
readonly specificationVersion = "v3";
readonly supportsStructuredOutputs: boolean;
readonly modelId: OpenAICompatibleChatModelId;
private readonly config;
private readonly failedResponseHandler;
private readonly chunkSchema;
constructor(modelId: OpenAICompatibleChatModelId, config: OpenAICompatibleChatConfig);
get provider(): string;
private get providerOptionsName();
get supportedUrls(): Record<string, RegExp[]> | PromiseLike<Record<string, RegExp[]>>;
private getArgs;
doGenerate(options: LanguageModelV3CallOptions): Promise<LanguageModelV3GenerateResult>;
doStream(options: LanguageModelV3CallOptions): Promise<LanguageModelV3StreamResult>;
}
type OpenAICompatibleCompletionModelId = string;
declare const openaiCompatibleCompletionProviderOptions: z.ZodObject<{
echo: z.ZodOptional<z.ZodBoolean>;
logitBias: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodNumber>>;
suffix: z.ZodOptional<z.ZodString>;
user: z.ZodOptional<z.ZodString>;
}, z.core.$strip>;
type OpenAICompatibleCompletionProviderOptions = z.infer<typeof openaiCompatibleCompletionProviderOptions>;
type OpenAICompatibleCompletionConfig = {
provider: string;
includeUsage?: boolean;
headers: () => Record<string, string | undefined>;
url: (options: {
modelId: string;
path: string;
}) => string;
fetch?: FetchFunction;
errorStructure?: ProviderErrorStructure<any>;
/**
* The supported URLs for the model.
*/
supportedUrls?: () => LanguageModelV3['supportedUrls'];
};
declare class OpenAICompatibleCompletionLanguageModel implements LanguageModelV3 {
readonly specificationVersion = "v3";
readonly modelId: OpenAICompatibleCompletionModelId;
private readonly config;
private readonly failedResponseHandler;
private readonly chunkSchema;
constructor(modelId: OpenAICompatibleCompletionModelId, config: OpenAICompatibleCompletionConfig);
get provider(): string;
private get providerOptionsName();
get supportedUrls(): Record<string, RegExp[]> | PromiseLike<Record<string, RegExp[]>>;
private getArgs;
doGenerate(options: LanguageModelV3CallOptions): Promise<LanguageModelV3GenerateResult>;
doStream(options: LanguageModelV3CallOptions): Promise<LanguageModelV3StreamResult>;
}
type OpenAICompatibleEmbeddingModelId = string;
declare const openaiCompatibleEmbeddingProviderOptions: z.ZodObject<{
dimensions: z.ZodOptional<z.ZodNumber>;
user: z.ZodOptional<z.ZodString>;
}, z.core.$strip>;
---
## Assistant
Tool: bash
Input:
{
"command": "curl -sL \"https://unpkg.com/@ai-sdk/openai-compatible@latest/dist/index.d.ts\" | grep -E \"chat|languageModel|chatModel\" | head -30",
"description": "Check if openai-compatible has chat method"
}Output:
languageModel(modelId: CHAT_MODEL_IDS, config?: Partial<OpenAICompatibleChatConfig>): LanguageModelV3;
chatModel(modelId: CHAT_MODEL_IDS): LanguageModelV3;
* Whether the provider supports structured outputs in chat models.
---
## Assistant
The `@ai-sdk/openai-compatible` package does have a `chatModel` method but **not** `chat`. The custom loader at line 373-374 calls `sdk.chat(modelID)` which doesn't exist!
Let me verify this is indeed the issue and prepare a fix:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 356,
"limit": 40
}Output:
<file>
00357| const accountId = Env.get("CLOUDFLARE_ACCOUNT_ID")
00358| const gateway = Env.get("CLOUDFLARE_GATEWAY_ID")
00359|
00360| if (!accountId || !gateway) return { autoload: false }
00361|
00362| // Get API token from env or auth prompt
00363| const apiToken = await (async () => {
00364| const envToken = Env.get("CLOUDFLARE_API_TOKEN")
00365| if (envToken) return envToken
00366| const auth = await Auth.get(input.id)
00367| if (auth?.type === "api") return auth.key
00368| return undefined
00369| })()
00370|
00371| return {
00372| autoload: true,
00373| async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
00374| return sdk.chat(modelID)
00375| },
00376| options: {
00377| baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
00378| headers: {
00379| // Cloudflare AI Gateway uses cf-aig-authorization for authenticated gateways
00380| // This enables Unified Billing where Cloudflare handles upstream provider auth
00381| ...(apiToken ? { "cf-aig-authorization": `Bearer ${apiToken}` } : {}),
00382| "HTTP-Referer": "https://opencode.ai/",
00383| "X-Title": "opencode",
00384| },
00385| // Custom fetch to strip Authorization header - AI Gateway uses cf-aig-authorization instead
00386| // Sending Authorization header with invalid value causes auth errors
00387| fetch: async (input: RequestInfo | URL, init?: RequestInit) => {
00388| const headers = new Headers(init?.headers)
00389| headers.delete("Authorization")
00390| return fetch(input, { ...init, headers })
00391| },
00392| },
00393| }
00394| },
00395| cerebras: async () => {
00396| return {
(File has more lines. Use 'offset' parameter to read beyond line 396)
</file>
---
## Assistant
Now I see the problem clearly. The `cloudflare-ai-gateway` custom loader has a `getModel` function that calls `sdk.chat(modelID)`, but the `@ai-sdk/openai-compatible` SDK doesn't have a `.chat()` method - it has `chatModel()` or `languageModel()`.
Let me check how other providers using `@ai-sdk/openai-compatible` handle this:
Tool: grep
Input:
{
"pattern": "openai-compatible",
"path": "/Users/englishm/git/opencode/packages/opencode/src/provider"
}Output:
Found 10 matches
/Users/englishm/git/opencode/packages/opencode/src/provider/sdk/openai-compatible/src/openai-compatible-provider.ts:
Line 2: import { OpenAICompatibleChatLanguageModel } from "@ai-sdk/openai-compatible"
Line 66: const getHeaders = () => withUserAgentSuffix(headers, `ai-sdk/openai-compatible/${VERSION}`)
Line 70: provider: `${options.name ?? "openai-compatible"}.chat`,
Line 79: provider: `${options.name ?? "openai-compatible"}.responses`,
/Users/englishm/git/opencode/packages/opencode/src/provider/sdk/openai-compatible/src/index.ts:
Line 1: export { createOpenaiCompatible, openaiCompatible } from "./openai-compatible-provider"
Line 2: export type { OpenaiCompatibleProvider, OpenaiCompatibleProviderSettings } from "./openai-compatible-provider"
/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts:
Line 25: import { createOpenAICompatible } from "@ai-sdk/openai-compatible"
Line 27: import { createOpenaiCompatible as createGitHubCopilotOpenAICompatible } from "./sdk/openai-compatible/src"
Line 49: "@ai-sdk/openai-compatible": createOpenAICompatible,
Line 839: if (model.api.npm.includes("@ai-sdk/openai-compatible") && options["includeUsage"] !== false) {
---
## Assistant
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/sdk/openai-compatible/src/openai-compatible-provider.ts"
}Output:
<file>
00001| import type { LanguageModelV2 } from "@ai-sdk/provider"
00002| import { OpenAICompatibleChatLanguageModel } from "@ai-sdk/openai-compatible"
00003| import { type FetchFunction, withoutTrailingSlash, withUserAgentSuffix } from "@ai-sdk/provider-utils"
00004| import { OpenAIResponsesLanguageModel } from "./responses/openai-responses-language-model"
00005|
00006| // Import the version or define it
00007| const VERSION = "0.1.0"
00008|
00009| export type OpenaiCompatibleModelId = string
00010|
00011| export interface OpenaiCompatibleProviderSettings {
00012| /**
00013| * API key for authenticating requests.
00014| */
00015| apiKey?: string
00016|
00017| /**
00018| * Base URL for the OpenAI Compatible API calls.
00019| */
00020| baseURL?: string
00021|
00022| /**
00023| * Name of the provider.
00024| */
00025| name?: string
00026|
00027| /**
00028| * Custom headers to include in the requests.
00029| */
00030| headers?: Record<string, string>
00031|
00032| /**
00033| * Custom fetch implementation.
00034| */
00035| fetch?: FetchFunction
00036| }
00037|
00038| export interface OpenaiCompatibleProvider {
00039| (modelId: OpenaiCompatibleModelId): LanguageModelV2
00040| chat(modelId: OpenaiCompatibleModelId): LanguageModelV2
00041| responses(modelId: OpenaiCompatibleModelId): LanguageModelV2
00042| languageModel(modelId: OpenaiCompatibleModelId): LanguageModelV2
00043|
00044| // embeddingModel(modelId: any): EmbeddingModelV2
00045|
00046| // imageModel(modelId: any): ImageModelV2
00047| }
00048|
00049| /**
00050| * Create an OpenAI Compatible provider instance.
00051| */
00052| export function createOpenaiCompatible(options: OpenaiCompatibleProviderSettings = {}): OpenaiCompatibleProvider {
00053| const baseURL = withoutTrailingSlash(options.baseURL ?? "https://api.openai.com/v1")
00054|
00055| if (!baseURL) {
00056| throw new Error("baseURL is required")
00057| }
00058|
00059| // Merge headers: defaults first, then user overrides
00060| const headers = {
00061| // Default OpenAI Compatible headers (can be overridden by user)
00062| ...(options.apiKey && { Authorization: `Bearer ${options.apiKey}` }),
00063| ...options.headers,
00064| }
00065|
00066| const getHeaders = () => withUserAgentSuffix(headers, `ai-sdk/openai-compatible/${VERSION}`)
00067|
00068| const createChatModel = (modelId: OpenaiCompatibleModelId) => {
00069| return new OpenAICompatibleChatLanguageModel(modelId, {
00070| provider: `${options.name ?? "openai-compatible"}.chat`,
00071| headers: getHeaders,
00072| url: ({ path }) => `${baseURL}${path}`,
00073| fetch: options.fetch,
00074| })
00075| }
00076|
00077| const createResponsesModel = (modelId: OpenaiCompatibleModelId) => {
00078| return new OpenAIResponsesLanguageModel(modelId, {
00079| provider: `${options.name ?? "openai-compatible"}.responses`,
00080| headers: getHeaders,
00081| url: ({ path }) => `${baseURL}${path}`,
00082| fetch: options.fetch,
00083| })
00084| }
00085|
00086| const createLanguageModel = (modelId: OpenaiCompatibleModelId) => createChatModel(modelId)
00087|
00088| const provider = function (modelId: OpenaiCompatibleModelId) {
00089| return createChatModel(modelId)
00090| }
00091|
00092| provider.languageModel = createLanguageModel
00093| provider.chat = createChatModel
00094| provider.responses = createResponsesModel
00095|
00096| return provider as OpenaiCompatibleProvider
00097| }
00098|
00099| // Default OpenAI Compatible provider instance
00100| export const openaiCompatible = createOpenaiCompatible()
00101|
(End of file - total 101 lines)
</file>
---
## Assistant
Now I see! OpenCode has its **own** custom `createOpenaiCompatible` wrapper at `packages/opencode/src/provider/sdk/openai-compatible/src/openai-compatible-provider.ts` that **does** have a `.chat()` method.
But the issue is that the bundled providers map at line 49 uses `createOpenAICompatible` from `@ai-sdk/openai-compatible` directly, NOT the custom one:
```typescript
"@ai-sdk/openai-compatible": createOpenAICompatible, // From @ai-sdk/openai-compatible - no .chat() method!
Let me verify this is the root cause:
Tool: read
**Input:**
```json
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 20,
"limit": 35
}
Output:
<file>
00021| import { createGoogleGenerativeAI } from "@ai-sdk/google"
00022| import { createVertex } from "@ai-sdk/google-vertex"
00023| import { createVertexAnthropic } from "@ai-sdk/google-vertex/anthropic"
00024| import { createOpenAI } from "@ai-sdk/openai"
00025| import { createOpenAICompatible } from "@ai-sdk/openai-compatible"
00026| import { createOpenRouter, type LanguageModelV2 } from "@openrouter/ai-sdk-provider"
00027| import { createOpenaiCompatible as createGitHubCopilotOpenAICompatible } from "./sdk/openai-compatible/src"
00028| import { createXai } from "@ai-sdk/xai"
00029| import { createMistral } from "@ai-sdk/mistral"
00030| import { createGroq } from "@ai-sdk/groq"
00031| import { createDeepInfra } from "@ai-sdk/deepinfra"
00032| import { createCerebras } from "@ai-sdk/cerebras"
00033| import { createCohere } from "@ai-sdk/cohere"
00034| import { createGateway } from "@ai-sdk/gateway"
00035| import { createTogetherAI } from "@ai-sdk/togetherai"
00036| import { createPerplexity } from "@ai-sdk/perplexity"
00037|
00038| export namespace Provider {
00039| const log = Log.create({ service: "provider" })
00040|
00041| const BUNDLED_PROVIDERS: Record<string, (options: any) => SDK> = {
00042| "@ai-sdk/amazon-bedrock": createAmazonBedrock,
00043| "@ai-sdk/anthropic": createAnthropic,
00044| "@ai-sdk/azure": createAzure,
00045| "@ai-sdk/google": createGoogleGenerativeAI,
00046| "@ai-sdk/google-vertex": createVertex,
00047| "@ai-sdk/google-vertex/anthropic": createVertexAnthropic,
00048| "@ai-sdk/openai": createOpenAI,
00049| "@ai-sdk/openai-compatible": createOpenAICompatible,
00050| "@openrouter/ai-sdk-provider": createOpenRouter,
00051| "@ai-sdk/xai": createXai,
00052| "@ai-sdk/mistral": createMistral,
00053| "@ai-sdk/groq": createGroq,
00054| "@ai-sdk/deepinfra": createDeepInfra,
00055| "@ai-sdk/cerebras": createCerebras,
(File has more lines. Use 'offset' parameter to read beyond line 55)
</file>
---
## Assistant
**Found the issue!**
The problem is:
1. `cloudflare-ai-gateway` uses `"npm": "@ai-sdk/openai-compatible"` from models.dev
2. The bundled provider at line 49 maps to `createOpenAICompatible` from `@ai-sdk/openai-compatible`
3. The `@ai-sdk/openai-compatible` SDK uses `chatModel()` or `languageModel()` methods, **NOT** `.chat()`
4. But the custom loader at lines 373-374 calls `sdk.chat(modelID)` which doesn't exist on the SDK
The fix should either:
1. Change the custom loader to use `sdk.languageModel(modelID)` instead of `sdk.chat(modelID)`, OR
2. Use the custom `createGitHubCopilotOpenAICompatible` for cloudflare-ai-gateway (which has `.chat()`)
Let me check which approach is more appropriate by checking the official SDK's API:
Tool: bash
Input:
{
"command": "curl -sL \"https://unpkg.com/@ai-sdk/openai-compatible@latest/dist/index.d.ts\" | grep -A5 \"interface.*Provider\"",
"description": "Check openai-compatible Provider interface"
}Output:
interface OpenAICompatibleProvider<CHAT_MODEL_IDS extends string = string, COMPLETION_MODEL_IDS extends string = string, EMBEDDING_MODEL_IDS extends string = string, IMAGE_MODEL_IDS extends string = string> extends Omit<ProviderV3, 'imageModel'> {
(modelId: CHAT_MODEL_IDS): LanguageModelV3;
languageModel(modelId: CHAT_MODEL_IDS, config?: Partial<OpenAICompatibleChatConfig>): LanguageModelV3;
chatModel(modelId: CHAT_MODEL_IDS): LanguageModelV3;
completionModel(modelId: COMPLETION_MODEL_IDS): LanguageModelV3;
embeddingModel(modelId: EMBEDDING_MODEL_IDS): EmbeddingModelV3;
--
interface OpenAICompatibleProviderSettings {
/**
Base URL for the API calls.
*/
baseURL: string;
/**
---
## Assistant
The official `@ai-sdk/openai-compatible` SDK has `languageModel()` and `chatModel()` but **NOT** `.chat()`.
The simplest fix is to change `sdk.chat(modelID)` to `sdk.languageModel(modelID)` in the cloudflare-ai-gateway custom loader:
Tool: todowrite
Input:
{
"todos": [
{
"id": "1",
"content": "Fix cloudflare-ai-gateway: change sdk.chat() to sdk.languageModel()",
"status": "in_progress",
"priority": "high"
}
]
}Output:
[
{
"id": "1",
"content": "Fix cloudflare-ai-gateway: change sdk.chat() to sdk.languageModel()",
"status": "in_progress",
"priority": "high"
}
]
---
## Assistant
Tool: edit
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"oldString": " return {\n autoload: true,\n async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {\n return sdk.chat(modelID)\n },\n options: {\n baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,",
"newString": " return {\n autoload: true,\n async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {\n return sdk.languageModel(modelID)\n },\n options: {\n baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,"
}
---
## Assistant
Now let me verify the fix is applied correctly:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/src/provider/provider.ts",
"offset": 369,
"limit": 10
}Output:
<file>
00370|
00371| return {
00372| autoload: true,
00373| async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
00374| return sdk.languageModel(modelID)
00375| },
00376| options: {
00377| baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
00378| headers: {
00379| // Cloudflare AI Gateway uses cf-aig-authorization for authenticated gateways
(File has more lines. Use 'offset' parameter to read beyond line 379)
</file>
---
## Assistant
Let me also check if there are any tests for this provider that need updating:
Tool: grep
Input:
{
"pattern": "cloudflare-ai-gateway",
"path": "/Users/englishm/git/opencode/packages/opencode/test"
}Output:
No files found
---
## Assistant
Tool: bash
Input:
{
"command": "cd packages/opencode && bun run build 2>&1 | head -50",
"description": "Build to check for errors",
"timeout": 120000
}Output:
/bin/bash: bun: command not found
---
## Assistant
Tool: bash
Input:
{
"command": "which bun || echo \"bun not found\"",
"description": "Check if bun is available"
}Output:
bun not found
---
## Assistant
Tool: bash
Input:
{
"command": "npx tsc --noEmit -p packages/opencode/tsconfig.json 2>&1 | head -30",
"description": "Type check the fix",
"timeout": 120000
}Output:
npm warn exec The following package was not found and will be installed: tsc@2.0.4
�[41m �[0m
�[41m�[37m This is not the tsc command you are looking for �[0m
�[41m �[0m
To get access to the TypeScript compiler, �[34mtsc�[0m, from the command line either:
- Use �[1mnpm install typescript�[0m to first add TypeScript to your project �[1mbefore�[0m using npx
- Use �[1myarn�[0m to avoid accidentally running code from un-installed packages
---
## Assistant
Tool: bash
Input:
{
"command": "~/.bun/bin/bun run build 2>&1 | head -50",
"description": "Build to check for errors",
"timeout": 120000,
"workdir": "/Users/englishm/git/opencode/packages/opencode"
}Output:
/bin/bash: /Users/englishm/.bun/bin/bun: No such file or directory
---
## Assistant
Tool: bash
Input:
{
"command": "ls -la /opt/homebrew/bin/bun 2>/dev/null || ls -la /usr/local/bin/bun 2>/dev/null || find /opt -name \"bun\" -type f 2>/dev/null | head -5",
"description": "Find bun executable"
}
---
## Assistant
The build environment isn't set up, but the fix is straightforward. Let me update the todo and provide a summary:
Tool: todowrite
Input:
{
"todos": [
{
"id": "1",
"content": "Fix cloudflare-ai-gateway: change sdk.chat() to sdk.languageModel()",
"status": "completed",
"priority": "high"
}
]
}Output:
[
{
"id": "1",
"content": "Fix cloudflare-ai-gateway: change sdk.chat() to sdk.languageModel()",
"status": "completed",
"priority": "high"
}
]
---
## Assistant
## Summary
I found and fixed the issue. The problem was in `/packages/opencode/src/provider/provider.ts` at line 374.
**Root Cause:**
The `cloudflare-ai-gateway` custom loader was calling `sdk.chat(modelID)`, but the `@ai-sdk/openai-compatible` SDK (which is used for this provider per models.dev) doesn't have a `.chat()` method - it only has `.languageModel()` and `.chatModel()`.
**Fix:**
Changed `sdk.chat(modelID)` to `sdk.languageModel(modelID)` in the cloudflare-ai-gateway custom loader.
This is the same type of issue as #4977 (GitHub Copilot Enterprise), where the wrong SDK method was being called. The difference is that #4977 was fixed by ensuring the correct npm package was assigned, whereas this fix corrects the method call to match the actual SDK API.
Would you like me to create a commit for this fix?
---
## User
Sorry, my dev environment on this machine isn't totally set up. I can install a few more things if necessary. I'm also curious if you can figure out what changed to break this because there's at least some evidence that people were able to use this provider at some point. What changed that led this to break and is our fix the correct one in light of whatever change broke things? Finally, I'm curious if I can run opencode directly from this repo with our fixes to validate with my own account when we're done and ready to open a PR. Thanks!
---
## Assistant
Great questions! Let me investigate the history to understand what changed and broke this.
Tool: bash
Input:
{
"command": "git log --oneline --all -20 -- packages/opencode/src/provider/provider.ts",
"description": "View recent changes to provider.ts"
}Output:
896d18ab3 chore: generate
893888536 fix(bedrock): support region and bearer token configuration (#6332)
1f2a4537b providers: add Cloudflare AI Gateway (#5174)
2fd23715f fix(providers.opencode): check config for api key in addition to auth (#5906)
141d24c28 fix: bundle more providers to fix breaking ai sdk issue
238ca62ba wip(desktop): progress
262d836dd LLM cleanup (#5462)
9898fbe8e providers: add Cloudflare AI Gateway (#5174)
87b5b3428 fix(providers.opencode): check config for api key in addition to auth (#5906)
f9be2bab3 fix: bundle more providers to fix breaking ai sdk issue
4a8e8f537 wip(desktop): progress
fed477645 LLM cleanup (#5462)
cf83f59dc Merge branch 'dev' into llm-centralization
df55ad89a Add Cerebras integration header with opencode identifier (#5354)
ffc4d5392 add models.dev family
915559b53 Merge branch 'dev' into llm-centralization
df64612d5 better interleaved thinking support (#5298)
8f3f87943 fmt
7ac3f9182 fixes
366b6e4fb wip
---
## Assistant
Tool: bash
Input:
{
"command": "git show 1f2a4537b --stat",
"description": "View the Cloudflare AI Gateway PR"
}Output:
commit 1f2a4537bcd969454dd7f33e14e4b20c3a6a7e78
Author: Matt Silverlock <matt@eatsleeprepeat.net>
Date: Tue Dec 23 18:31:58 2025 -0500
providers: add Cloudflare AI Gateway (#5174)
packages/opencode/src/provider/provider.ts | 39 +++++++++++++++++++
packages/web/src/content/docs/providers.mdx | 58 +++++++++++++++++++++++++++++
2 files changed, 97 insertions(+)
---
## Assistant
Tool: bash
Input:
{
"command": "git show 1f2a4537b -- packages/opencode/src/provider/provider.ts",
"description": "View original Cloudflare implementation"
}Output:
commit 1f2a4537bcd969454dd7f33e14e4b20c3a6a7e78
Author: Matt Silverlock <matt@eatsleeprepeat.net>
Date: Tue Dec 23 18:31:58 2025 -0500
providers: add Cloudflare AI Gateway (#5174)
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index b11ca9368..0fdf26392 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -338,6 +338,45 @@ export namespace Provider {
},
}
},
+ "cloudflare-ai-gateway": async (input) => {
+ const accountId = Env.get("CLOUDFLARE_ACCOUNT_ID")
+ const gateway = Env.get("CLOUDFLARE_GATEWAY_ID")
+
+ if (!accountId || !gateway) return { autoload: false }
+
+ // Get API token from env or auth prompt
+ const apiToken = await (async () => {
+ const envToken = Env.get("CLOUDFLARE_API_TOKEN")
+ if (envToken) return envToken
+ const auth = await Auth.get(input.id)
+ if (auth?.type === "api") return auth.key
+ return undefined
+ })()
+
+ return {
+ autoload: true,
+ async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+ return sdk.chat(modelID)
+ },
+ options: {
+ baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
+ headers: {
+ // Cloudflare AI Gateway uses cf-aig-authorization for authenticated gateways
+ // This enables Unified Billing where Cloudflare handles upstream provider auth
+ ...(apiToken ? { "cf-aig-authorization": `Bearer ${apiToken}` } : {}),
+ "HTTP-Referer": "https://opencode.ai/",
+ "X-Title": "opencode",
+ },
+ // Custom fetch to strip Authorization header - AI Gateway uses cf-aig-authorization instead
+ // Sending Authorization header with invalid value causes auth errors
+ fetch: async (input: RequestInfo | URL, init?: RequestInit) => {
+ const headers = new Headers(init?.headers)
+ headers.delete("Authorization")
+ return fetch(input, { ...init, headers })
+ },
+ },
+ }
+ },
cerebras: async () => {
return {
autoload: false,
---
## Assistant
So the original implementation already had `sdk.chat(modelID)`. Let me check if there was a different SDK being used before, or if the bundled providers changed:
Tool: bash
Input:
{
"command": "git log --oneline --all -30 -- packages/opencode/src/provider/provider.ts | xargs -I {} sh -c 'echo \"=== {} ===\" && git show {} -- packages/opencode/src/provider/provider.ts 2>/dev/null | grep -A3 \"BUNDLED_PROVIDERS\" | head -20'",
"description": "Check BUNDLED_PROVIDERS history"
}Output:
=== 896d18ab3 chore: generate ===
xargs: command line cannot be assembled, too long
---
## Assistant
Tool: bash
Input:
{
"command": "git log --oneline -10 -- packages/opencode/src/provider/provider.ts",
"description": "View recent commits"
}Output:
896d18ab3 chore: generate
893888536 fix(bedrock): support region and bearer token configuration (#6332)
9898fbe8e providers: add Cloudflare AI Gateway (#5174)
87b5b3428 fix(providers.opencode): check config for api key in addition to auth (#5906)
f9be2bab3 fix: bundle more providers to fix breaking ai sdk issue
4a8e8f537 wip(desktop): progress
fed477645 LLM cleanup (#5462)
df55ad89a Add Cerebras integration header with opencode identifier (#5354)
ffc4d5392 add models.dev family
df64612d5 better interleaved thinking support (#5298)
---
## Assistant
Tool: bash
Input:
{
"command": "git show f9be2bab3 -- packages/opencode/src/provider/provider.ts | head -100",
"description": "View \"bundle more providers\" fix"
}Output:
commit f9be2bab3af7cee47d6b03d00938570dad6ad541
Author: Aiden Cline <aidenpcline@gmail.com>
Date: Mon Dec 22 15:12:18 2025 -0600
fix: bundle more providers to fix breaking ai sdk issue
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index b8d4dadbd..5e4464339 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -25,6 +25,15 @@ import { createOpenAI } from "@ai-sdk/openai"
import { createOpenAICompatible } from "@ai-sdk/openai-compatible"
import { createOpenRouter, type LanguageModelV2 } from "@openrouter/ai-sdk-provider"
import { createOpenaiCompatible as createGitHubCopilotOpenAICompatible } from "./sdk/openai-compatible/src"
+import { createXai } from "@ai-sdk/xai"
+import { createMistral } from "@ai-sdk/mistral"
+import { createGroq } from "@ai-sdk/groq"
+import { createDeepInfra } from "@ai-sdk/deepinfra"
+import { createCerebras } from "@ai-sdk/cerebras"
+import { createCohere } from "@ai-sdk/cohere"
+import { createGateway } from "@ai-sdk/gateway"
+import { createTogetherAI } from "@ai-sdk/togetherai"
+import { createPerplexity } from "@ai-sdk/perplexity"
export namespace Provider {
const log = Log.create({ service: "provider" })
@@ -39,6 +48,15 @@ export namespace Provider {
"@ai-sdk/openai": createOpenAI,
"@ai-sdk/openai-compatible": createOpenAICompatible,
"@openrouter/ai-sdk-provider": createOpenRouter,
+ "@ai-sdk/xai": createXai,
+ "@ai-sdk/mistral": createMistral,
+ "@ai-sdk/groq": createGroq,
+ "@ai-sdk/deepinfra": createDeepInfra,
+ "@ai-sdk/cerebras": createCerebras,
+ "@ai-sdk/cohere": createCohere,
+ "@ai-sdk/gateway": createGateway,
+ "@ai-sdk/togetherai": createTogetherAI,
+ "@ai-sdk/perplexity": createPerplexity,
// @ts-ignore (TODO: kill this code so we dont have to maintain it)
"@ai-sdk/github-copilot": createGitHubCopilotOpenAICompatible,
}
---
## Assistant
Let me check the commit before the Cloudflare addition to see what SDK was being used:
Tool: bash
Input:
{
"command": "git show 9898fbe8e^:packages/opencode/src/provider/provider.ts | grep -A20 \"BUNDLED_PROVIDERS\"",
"description": "View BUNDLED_PROVIDERS before Cloudflare"
}Output:
const BUNDLED_PROVIDERS: Record<string, (options: any) => SDK> = {
"@ai-sdk/amazon-bedrock": createAmazonBedrock,
"@ai-sdk/anthropic": createAnthropic,
"@ai-sdk/azure": createAzure,
"@ai-sdk/google": createGoogleGenerativeAI,
"@ai-sdk/google-vertex": createVertex,
"@ai-sdk/google-vertex/anthropic": createVertexAnthropic,
"@ai-sdk/openai": createOpenAI,
"@ai-sdk/openai-compatible": createOpenAICompatible,
"@openrouter/ai-sdk-provider": createOpenRouter,
"@ai-sdk/xai": createXai,
"@ai-sdk/mistral": createMistral,
"@ai-sdk/groq": createGroq,
"@ai-sdk/deepinfra": createDeepInfra,
"@ai-sdk/cerebras": createCerebras,
"@ai-sdk/cohere": createCohere,
"@ai-sdk/gateway": createGateway,
"@ai-sdk/togetherai": createTogetherAI,
"@ai-sdk/perplexity": createPerplexity,
// @ts-ignore (TODO: kill this code so we dont have to maintain it)
"@ai-sdk/github-copilot": createGitHubCopilotOpenAICompatible,
--
const bundledFn = BUNDLED_PROVIDERS[bundledKey]
if (bundledFn) {
log.info("using bundled provider", { providerID: model.providerID, pkg: bundledKey })
const loaded = bundledFn({
name: model.providerID,
...options,
})
s.sdk.set(key, loaded)
return loaded as SDK
}
let installedPath: string
if (!model.api.npm.startsWith("file://")) {
installedPath = await BunProc.install(model.api.npm, "latest")
} else {
log.info("loading local provider", { pkg: model.api.npm })
installedPath = model.api.npm
}
const mod = await import(installedPath)
---
## Assistant
Now let me check if the `@ai-sdk/openai-compatible` package API changed. Let me look at the package history:
Tool: bash
Input:
{
"command": "git log --oneline -5 -- packages/opencode/src/provider/sdk/openai-compatible",
"description": "Check custom SDK history"
}Output:
6e2379a28 get codex working in copilot (#4914)
---
## Assistant
Tool: bash
Input:
{
"command": "git show 6e2379a28 --stat",
"description": "View codex/copilot fix"
}Output:
commit 6e2379a28cf434009276007f383528e3ebe4ac02
Author: Aiden Cline <63023139+rekram1-node@users.noreply.github.com>
Date: Sat Nov 29 19:45:33 2025 -0800
get codex working in copilot (#4914)
Co-authored-by: OpeOginni <107570612+OpeOginni@users.noreply.github.com>
Co-authored-by: GitHub Action <action@github.com>
bun.lock | 16 +-
flake.lock | 6 +-
nix/hashes.json | 2 +-
packages/opencode/package.json | 2 +
.../src/cli/cmd/tui/component/prompt/index.tsx | 1 +
packages/opencode/src/provider/provider.ts | 48 +-
.../provider/sdk/openai-compatible/src/README.md | 5 +
.../provider/sdk/openai-compatible/src/index.ts | 2 +
.../src/openai-compatible-provider.ts | 100 ++
.../responses/convert-to-openai-responses-input.ts | 303 ++++
.../map-openai-responses-finish-reason.ts | 22 +
.../src/responses/openai-config.ts | 18 +
.../src/responses/openai-error.ts | 22 +
.../src/responses/openai-responses-api-types.ts | 207 +++
.../responses/openai-responses-language-model.ts | 1713 ++++++++++++++++++++
.../responses/openai-responses-prepare-tools.ts | 177 ++
.../src/responses/openai-responses-settings.ts | 1 +
.../src/responses/tool/code-interpreter.ts | 88 +
.../src/responses/tool/file-search.ts | 128 ++
.../src/responses/tool/image-generation.ts | 115 ++
.../src/responses/tool/local-shell.ts | 65 +
.../src/responses/tool/web-search-preview.ts | 104 ++
.../src/responses/tool/web-search.ts | 103 ++
packages/plugin/package.json | 2 +-
packages/sdk/js/package.json | 2 +-
25 files changed, 3230 insertions(+), 22 deletions(-)
---
## Assistant
Tool: bash
Input:
{
"command": "git show 6e2379a28 -- packages/opencode/src/provider/provider.ts | head -80",
"description": "View the codex copilot changes"
}Output:
commit 6e2379a28cf434009276007f383528e3ebe4ac02
Author: Aiden Cline <63023139+rekram1-node@users.noreply.github.com>
Date: Sat Nov 29 19:45:33 2025 -0800
get codex working in copilot (#4914)
Co-authored-by: OpeOginni <107570612+OpeOginni@users.noreply.github.com>
Co-authored-by: GitHub Action <action@github.com>
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index ded0b9b19..9da4ca161 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -23,6 +23,7 @@ import { createVertexAnthropic } from "@ai-sdk/google-vertex/anthropic"
import { createOpenAI } from "@ai-sdk/openai"
import { createOpenAICompatible } from "@ai-sdk/openai-compatible"
import { createOpenRouter } from "@openrouter/ai-sdk-provider"
+import { createOpenaiCompatible as createGitHubCopilotOpenAICompatible } from "./sdk/openai-compatible/src"
export namespace Provider {
const log = Log.create({ service: "provider" })
@@ -37,6 +38,8 @@ export namespace Provider {
"@ai-sdk/openai": createOpenAI,
"@ai-sdk/openai-compatible": createOpenAICompatible,
"@openrouter/ai-sdk-provider": createOpenRouter,
+ // @ts-ignore (TODO: kill this code so we dont have to maintain it)
+ "@ai-sdk/github-copilot": createGitHubCopilotOpenAICompatible,
}
type CustomLoader = (provider: ModelsDev.Provider) => Promise<{
@@ -87,6 +90,30 @@ export namespace Provider {
options: {},
}
},
+ "github-copilot": async () => {
+ return {
+ autoload: false,
+ async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+ if (modelID.includes("gpt-5")) {
+ return sdk.responses(modelID)
+ }
+ return sdk.chat(modelID)
+ },
+ options: {},
+ }
+ },
+ "github-copilot-enterprise": async () => {
+ return {
+ autoload: false,
+ async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+ if (modelID.includes("gpt-5")) {
+ return sdk.responses(modelID)
+ }
+ return sdk.chat(modelID)
+ },
+ options: {},
+ }
+ },
azure: async () => {
return {
autoload: false,
@@ -428,15 +455,6 @@ export namespace Provider {
}
}
- // load custom
- for (const [providerID, fn] of Object.entries(CUSTOM_LOADERS)) {
- if (disabled.has(providerID)) continue
- const result = await fn(database[providerID])
- if (result && (result.autoload || providers[providerID])) {
- mergeProvider(providerID, result.options ?? {}, "custom", result.getModel)
- }
- }
-
for (const plugin of await Plugin.list()) {
if (!plugin.auth) continue
const providerID = plugin.auth.provider
@@ -478,6 +496,14 @@ export namespace Provider {
}
}
---
## Assistant
I see! The custom `createGitHubCopilotOpenAICompatible` was introduced for GitHub Copilot specifically, and it **does have** the `.chat()` method. This is the key finding.
Let me check when `@ai-sdk/openai-compatible` was updated to change its API:
Tool: bash
Input:
{
"command": "npm view @ai-sdk/openai-compatible versions --json 2>/dev/null | tail -20",
"description": "Check openai-compatible versions"
}Output:
"2.0.0-beta.44",
"2.0.0-beta.45",
"2.0.0-beta.46",
"2.0.0-beta.47",
"2.0.0-beta.48",
"2.0.0-beta.49",
"2.0.0-beta.50",
"2.0.0-beta.51",
"2.0.0-beta.52",
"2.0.0-beta.53",
"2.0.0-beta.54",
"2.0.0-beta.55",
"2.0.0-beta.56",
"2.0.0-beta.57",
"2.0.0-beta.58",
"2.0.0-beta.59",
"2.0.0-beta.60",
"2.0.0",
"2.0.1"
]
---
## Assistant
Tool: bash
Input:
{
"command": "npm view @ai-sdk/openai-compatible@0.1.0 2>/dev/null | head -20 || npm view @ai-sdk/openai-compatible time --json 2>/dev/null | head -30",
"description": "Check package version history"
}Output:
@ai-sdk/openai-compatible@0.1.0 | Apache-2.0 | deps: 2 | versions: 192
This package provides a foundation for implementing providers that expose an OpenAI-compatible API.
https://sdk.vercel.ai/docs
keywords: ai
dist
.tarball: https://registry.npmjs.org/@ai-sdk/openai-compatible/-/openai-compatible-0.1.0.tgz
.shasum: ec5f4c22a927a1b4499a6c2a4163ec60e7c56fe3
.integrity: sha512-sgk8oPdnTwImR77HT+esnjoZ7Uu4V5ptXJY+w06GGRCeM3PMJowAqC7gAdYhD5DBYUo+4Gs+wvKenaPBNV0WVg==
.unpackedSize: 285.7 kB
dependencies:
@ai-sdk/provider-utils: 2.1.0
@ai-sdk/provider: 1.0.4
maintainers:
- vercel-release-bot <infra+release@vercel.com>
- matheuss <matheus.frndes@gmail.com>
---
## Assistant
Tool: bash
Input:
{
"command": "curl -sL \"https://unpkg.com/@ai-sdk/openai-compatible@0.1.0/dist/index.d.ts\" | grep -E \"chat\\(|chatModel\\(|languageModel\\(\" | head -10",
"description": "Check old API of openai-compatible"
}Output:
languageModel(modelId: CHAT_MODEL_IDS, settings?: OpenAICompatibleChatSettings): LanguageModelV1;
chatModel(modelId: CHAT_MODEL_IDS, settings?: OpenAICompatibleChatSettings): LanguageModelV1;
---
## Assistant
Interesting! Even the old version (0.1.0) didn't have a `.chat()` method - it only had `.chatModel()` and `.languageModel()`. Let me check if this was ever actually working by looking at the PR that added it:
Tool: bash
Input:
{
"command": "gh pr view 5174 --repo sst/opencode --json body,comments,reviews 2>/dev/null | head -100",
"description": "View the Cloudflare PR discussion"
}Output:
{"body":"This PR:\r\n\r\n- [x] Adds Cloudflare's AI Gateway (https://developers.cloudflare.com/ai-gateway/) as a provider to opencode\r\n- [x] Updates the Providers (https://opencode.ai/docs/providers/) docs\r\n- [x] Links to the corresponding models.dev `cloudflare-ai-gateway` provider to autoload models - PR is here: https://github.com/sst/models.dev/pull/476 \r\n- [x] Validate model selection works\r\n\r\n","comments":[{"id":"IC_kwDOOiiGL87X09wa","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"Let me see what’s up with the format failure when I’m back tomorrow. ","createdAt":"2025-12-06T19:16:27Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3620985882","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87X1VOb","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"@elithrar You can ignore the ci failure, it'll autoformat anything not formatted after being merged anyways","createdAt":"2025-12-06T20:23:23Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3621082011","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87X1Vc1","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"This looks good to me, models.dev has been updated and given it seems this works for you, it's safe to merge but I'll hold off till you are ready","createdAt":"2025-12-06T20:24:13Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3621082933","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87X1dpz","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"Thanks! I’m seeing a few issues with some Workers AI models - want to debug\r\nthat before this goes live. Suspect the model naming format isn’t working.\r\n\r\nOn Sat, Dec 6, 2025 at 15:24 Aiden Cline ***@***.***> wrote:\r\n\r\n> *rekram1-node* left a comment (sst/opencode#5174)\r\n> <https://github.com/sst/opencode/pull/5174#issuecomment-3621082933>\r\n>\r\n> This looks good to me, models.dev has been updated and given it seems\r\n> this works for you, it's safe to merge but I'll hold off till you are ready\r\n>\r\n> —\r\n> Reply to this email directly, view it on GitHub\r\n> <https://github.com/sst/opencode/pull/5174#issuecomment-3621082933>, or\r\n> unsubscribe\r\n> <https://github.com/notifications/unsubscribe-auth/AAAEQ4BYNAA4C72PN2BBVNL4AM3QHAVCNFSM6AAAAACOH6K3SGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMMRRGA4DEOJTGM>\r\n> .\r\n> You are receiving this because you were mentioned.Message ID:\r\n> ***@***.***>\r\n>\r\n","createdAt":"2025-12-06T21:00:33Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[{"content":"THUMBS_UP","users":{"totalCount":1}}],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3621116531","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87ZhWkU","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"Fixed the merge conflict. Waiting on https://github.com/sst/models.dev/pull/514 to be merged in.","createdAt":"2025-12-13T13:09:45Z","includesCreatedEdit":true,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3649399060","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87ZiMz7","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"@elithrar merged","createdAt":"2025-12-13T16:59:13Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3649621243","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87bzzBq","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"Btw (and we prolly need a cleaner way of doing stuff) but since opencode is still using AI Sdk v5 and the dynamic provider loading just grabs latest, I think if it tries loading `ai-gateway-provider` it will fail.\r\n\r\nYou can add the package into packages/opencode/package.json and pin the version to v5 compat and then update packages/opencode/..../provider/provider.ts to load it (youll see the logic im referring to at the top of the file)","createdAt":"2025-12-23T19:42:39Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3687788650","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b0C45","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"Got it, great - let me do that.\r\n\r\ntesting this again but at some point probably need to support a\r\n—models-dev-endpoint flag for overriding with a localhost instance of\r\nmodels.dev too so it’s easier to validate e2e\r\n\r\nOn Tue, Dec 23, 2025 at 14:43 Aiden Cline ***@***.***> wrote:\r\n\r\n> *rekram1-node* left a comment (sst/opencode#5174)\r\n> <https://github.com/sst/opencode/pull/5174#issuecomment-3687788650>\r\n>\r\n> Btw (and we prolly need a cleaner way of doing stuff) but since opencode\r\n> is still using AI Sdk v5 and the dynamic provider loading just grabs\r\n> latest, I think if it tries loading ai-gateway-provider it will fail.\r\n>\r\n> You can add the package into packages/opencode/package.json and pin the\r\n> version to v5 compat and then update\r\n> packages/opencode/..../provider/provider.ts to load it (youll see the logic\r\n> im referring to at the top of the file)\r\n>\r\n> —\r\n> Reply to this email directly, view it on GitHub\r\n> <https://github.com/sst/opencode/pull/5174#issuecomment-3687788650>, or\r\n> unsubscribe\r\n> <https://github.com/notifications/unsubscribe-auth/AAAEQ4GBXIE3VIKZGMEZGCT4DGLMLAVCNFSM6AAAAACOH6K3SGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMOBXG44DQNRVGA>\r\n> .\r\n> You are receiving this because you were mentioned.Message ID:\r\n> ***@***.***>\r\n>\r\n","createdAt":"2025-12-23T20:10:32Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[{"content":"THUMBS_UP","users":{"totalCount":1}}],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3687853625","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b0F-G","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"Yeah that's a good idea","createdAt":"2025-12-23T20:16:21Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3687866246","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b0Rgg","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"merge + deploy https://github.com/sst/models.dev/pull/560 first ","createdAt":"2025-12-23T20:40:45Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3687913504","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b0SxC","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"models.dev pr has been merged + deployed","createdAt":"2025-12-23T20:43:45Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3687918658","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b0ldW","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"* re: v5 compat - this should be OK - `@ai-sdk/openai-compatible ` is pinned to `v1.0.27` (v5 compat) - ?","createdAt":"2025-12-23T21:19:36Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3687995222","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b0zo3","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"I still crash the TUI on any model selection for Cloudflare AI Gateway: \r\n\r\nThe stack trace isn't useful as it's stuck in the spinning/TUI vs. the underlying provider exception:\r\n\r\n```\r\nError: [Reconciler] Unknown component type: spinner\r\n at createElement (../../node_modules/.bun/@opentui+solid@0.0.0-20251211-4403a69a+897ebc1644d3edd0/node_modules/@opentui/solid/index.js:521:17)\r\n at children (src/cli/cmd/tui/component/prompt/index.tsx:874:24)\r\n at <anonymous> (../../node_modules/.bun/solid-js@1.9.10/node_modules/solid-js/dist/dev.js:1568:19)\r\n at runComputation (../../node_modules/.bun/solid-js@1.9.10/node_modules/solid-js/dist/dev.js:742:22)\r\n at updateComputation (../../node_modules/.bun/solid-js@1.9.10/node_modules/solid-js/dist/dev.js:724:3)\r\n at readSignal (../../node_modules/.bun/solid-js@1.9.10/node_modules/solid-js/dist/dev.js:660:67)\r\n at <anonymous> (../../node_modules/.bun/@opentui+solid@0.0.0-20251211-4403a69a+897ebc1644d3edd0/node_modules/@opentui/solid/index.js:148:62)\r\n at runComputation (../../node_modules/.bun/solid-js@1.9.10/node_modules/solid-js/dist/dev.js:742:22)\r\n at updateComputation (../../node_modules/.bun/solid-js@1.9.10/node_modules/solid-js/dist/dev.js:724:3)\r\n at createRenderEffect (../../node_modules/.bun/solid-js@1.9.10/node_modules/solid-js/dist/dev.js:240:75)...\r\n```","createdAt":"2025-12-23T21:49:00Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3688053303","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b1Q8n","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"OK, we have this working. I overrode the models based on the latest models.dev/api.json + fixed the way we call the `openai-compatible` SDK (`.languageModel`).\r\n\r\nLocal builds from this branch work as expected: \r\n\r\n```sh\r\n# Set environment variables\r\nexport CLOUDFLARE_ACCOUNT_ID=\"your-account-id\"\r\nexport CLOUDFLARE_GATEWAY_ID=\"your-gateway-name\"\r\nexport CLOUDFLARE_API_TOKEN=\"your-api-token\"\r\n\r\n# OpenAI via Cloudflare AI Gateway\r\n./dist/opencode-darwin-arm64/bin/opencode run --model \"cloudflare-ai-gateway/openai/gpt-4o\" \"explain what opencode does\"\r\n\r\n# Anthropic via Cloudflare AI Gateway \r\n./dist/opencode-darwin-arm64/bin/opencode run --model \"cloudflare-ai-gateway/anthropic/claude-sonnet-4-5\" \"write a hello world in rust\"\r\n\r\n# Workers AI\r\n./dist/opencode-darwin-arm64/bin/opencode run --model \"cloudflare-ai-gateway/workers-ai/@cf/qwen/qwen2.5-coder-32b-instruct\" \"fix the bug in main.ts\"\r\n\r\n# List available models\r\n./dist/opencode-darwin-arm64/bin/opencode models | grep cloudflare-ai-gateway\r\n```\r\n\r\n```sh\r\n➜ ./dist/opencode-darwin-arm64/bin/opencode run --model \"cloudflare-ai-gateway/anthropic/claude-sonnet-4-5\" \"write a hello world in rust\"\r\n\r\nI'll create a simple \"Hello, World!\" program in Rust for you.\r\n\r\n| Write packages/opencode/hello.rs\r\n\r\nI've created `hello.rs` with a simple Hello World program. You can run it with:\r\n\r\nrustc hello.rs && ./hello\r\n\r\nOr if you have `cargo` and want to run it directly:\r\n\r\nrustc hello.rs && ./hello\r\n```","createdAt":"2025-12-23T23:02:08Z","includesCreatedEdit":true,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3688173351","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b1ZbO","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"Thanks!","createdAt":"2025-12-23T23:31:56Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3688208078","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b841_","author":{"login":"ahosker"},"authorAssociation":"CONTRIBUTOR","body":"Is OpenCode Zen to be added to CloudFlare AI Gateway @elithrar /or/ with this added is their a better way to do this?\r\n<img width=\"911\" height=\"160\" alt=\"image\" src=\"https://github.com/user-attachments/assets/cd23e57f-38fc-462d-a4ec-bcffc0d10406\" />\r\n","createdAt":"2025-12-24T16:09:49Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3690171775","viewerDidAuthor":false},{"id":"IC_kwDOOiiGL87b858c","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"@ahosker right now you’d want to set it up as a custom provider. Can’t speak to whether we have plans otherwise. ","createdAt":"2025-12-24T16:12:53Z","includesCreatedEdit":false,"isMinimized":false,"minimizedReason":"","reactionGroups":[{"content":"THUMBS_UP","users":{"totalCount":1}}],"url":"https://github.com/sst/opencode/pull/5174#issuecomment-3690176284","viewerDidAuthor":false}],"reviews":[{"id":"PRR_kwDOOiiGL87XIVfF","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"","submittedAt":"2025-12-23T20:44:42Z","includesCreatedEdit":false,"reactionGroups":[],"state":"COMMENTED","commit":{"oid":"e8140da138ede9797f68286d35f7cdbca07b8802"}},{"id":"PRR_kwDOOiiGL87XIVzm","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"","submittedAt":"2025-12-23T20:45:33Z","includesCreatedEdit":false,"reactionGroups":[],"state":"COMMENTED","commit":{"oid":"e8140da138ede9797f68286d35f7cdbca07b8802"}},{"id":"PRR_kwDOOiiGL87XIpv9","author":{"login":"elithrar"},"authorAssociation":"CONTRIBUTOR","body":"","submittedAt":"2025-12-23T21:20:17Z","includesCreatedEdit":false,"reactionGroups":[],"state":"COMMENTED","commit":{"oid":"e8140da138ede9797f68286d35f7cdbca07b8802"}},{"id":"PRR_kwDOOiiGL87XI-oe","author":{"login":"rekram1-node"},"authorAssociation":"COLLABORATOR","body":"","submittedAt":"2025-12-23T22:03:23Z","includesCreatedEdit":false,"reactionGroups":[],"state":"COMMENTED","commit":{"oid":"e8140da138ede9797f68286d35f7cdbca07b8802"}}]}
---
## Assistant
This is very helpful. I found it:
> "OK, we have this working. I overrode the models based on the latest models.dev/api.json + **fixed the way we call the `openai-compatible` SDK (`.languageModel`)**."
So the original author (@elithrar) **did** fix it to use `.languageModel()` before merging! Let me check what happened:
Tool: bash
Input:
{
"command": "git log --oneline --all -- packages/opencode/src/provider/provider.ts | grep -i cloudflare",
"description": "Find all Cloudflare-related commits"
}Output:
1f2a4537b providers: add Cloudflare AI Gateway (#5174)
9898fbe8e providers: add Cloudflare AI Gateway (#5174)
---
## Assistant
Tool: bash
Input:
{
"command": "git show 9898fbe8e -- packages/opencode/src/provider/provider.ts | grep -A10 \"cloudflare-ai-gateway\"",
"description": "Check the dev branch version"
}Output:
+ "cloudflare-ai-gateway": async (input) => {
+ const accountId = Env.get("CLOUDFLARE_ACCOUNT_ID")
+ const gateway = Env.get("CLOUDFLARE_GATEWAY_ID")
+
+ if (!accountId || !gateway) return { autoload: false }
+
+ // Get API token from env or auth prompt
+ const apiToken = await (async () => {
+ const envToken = Env.get("CLOUDFLARE_API_TOKEN")
+ if (envToken) return envToken
+ const auth = await Auth.get(input.id)
---
## Assistant
Tool: bash
Input:
{
"command": "git show 9898fbe8e -- packages/opencode/src/provider/provider.ts | grep -A5 \"getModel\"",
"description": "Check getModel in dev branch"
}Output:
+ async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+ return sdk.chat(modelID)
+ },
+ options: {
+ baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
+ headers: {
---
## Assistant
Tool: bash
Input:
{
"command": "git log --oneline 9898fbe8e..1f2a4537b -- packages/opencode/src/provider/provider.ts 2>/dev/null || git log --oneline --all | head -50",
"description": "Check commits between the two"
}Output:
1f2a4537b providers: add Cloudflare AI Gateway (#5174)
2fd23715f fix(providers.opencode): check config for api key in addition to auth (#5906)
141d24c28 fix: bundle more providers to fix breaking ai sdk issue
238ca62ba wip(desktop): progress
262d836dd LLM cleanup (#5462)
---
## Assistant
Tool: bash
Input:
{
"command": "git show 1f2a4537b -- packages/opencode/src/provider/provider.ts | grep -A5 \"getModel\"",
"description": "Check getModel in main branch merge"
}Output:
+ async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+ return sdk.chat(modelID)
+ },
+ options: {
+ baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
+ headers: {
---
## Assistant
So both commits have `sdk.chat(modelID)`. Let me check the PR's actual commits:
Tool: bash
Input:
{
"command": "gh pr view 5174 --repo sst/opencode --json commits --jq '.commits[].oid' | head -20",
"description": "Get PR commit SHAs"
}Output:
f8910c6cad5383e0e0a9eac0a65a7df846602b22
e8140da138ede9797f68286d35f7cdbca07b8802
727a1f8f7a44fc7a89056b139fa905b3f5c9fb7a
---
## Assistant
Tool: bash
Input:
{
"command": "git show 727a1f8f7a44fc7a89056b139fa905b3f5c9fb7a -- packages/opencode/src/provider/provider.ts 2>/dev/null | grep -A5 \"getModel\" || echo \"Commit not found locally\"",
"description": "Check the final PR commit"
}Output:
Commit not found locally
---
## Assistant
Tool: bash
Input:
{
"command": "git fetch origin pull/5174/head:pr-5174 2>&1 && git show pr-5174 -- packages/opencode/src/provider/provider.ts | grep -A5 \"getModel\"",
"description": "Fetch and check the PR branch"
}Output:
From github.com:sst/opencode
* [new ref] refs/pull/5174/head -> pr-5174
---
## Assistant
Tool: bash
Input:
{
"command": "git show pr-5174 -- packages/opencode/src/provider/provider.ts | grep -A5 \"getModel\"",
"description": "Check getModel in the PR"
}
---
## Assistant
Tool: bash
Input:
{
"command": "git show pr-5174 -- packages/opencode/src/provider/provider.ts | grep -B2 -A10 \"cloudflare\"",
"description": "Check cloudflare in the PR"
}
---
## Assistant
Tool: bash
Input:
{
"command": "git log pr-5174 --oneline -5",
"description": "Check PR commits"
}Output:
727a1f8f7 Merge branch 'dev' into cloudflare-ai-gateway
1bd8e6171 ci: adam is not a full stack engineer
b6c07cb1b fix: remove desktop dup
83f23817c Update gitlab.mdx to use the 2.x component version (#6062)
23b1d7c75 Update Nix flake.lock and hashes
---
## Assistant
Tool: bash
Input:
{
"command": "git log pr-5174 --oneline -20 | grep -i cloudflare",
"description": "Find cloudflare commits in PR"
}Output:
727a1f8f7 Merge branch 'dev' into cloudflare-ai-gateway
e8140da13 Merge branch 'dev' into cloudflare-ai-gateway
f8910c6ca feat(cloudflare-ai-gateway): add Cloudflare AI Gateway provider with Unified Billing support
---
## Assistant
Tool: bash
Input:
{
"command": "git show f8910c6ca -- packages/opencode/src/provider/provider.ts | grep -A10 \"getModel\"",
"description": "Check original cloudflare commit"
}Output:
- Add getModel function using sdk.chat() for OpenAI-compatible endpoint
- Support API token from env var or /connect command
- Update docs with setup instructions
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index b11ca9368..0fdf26392 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -338,6 +338,45 @@ export namespace Provider {
},
}
--
+ async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+ return sdk.chat(modelID)
+ },
+ options: {
+ baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
+ headers: {
+ // Cloudflare AI Gateway uses cf-aig-authorization for authenticated gateways
+ // This enables Unified Billing where Cloudflare handles upstream provider auth
+ ...(apiToken ? { "cf-aig-authorization": `Bearer ${apiToken}` } : {}),
+ "HTTP-Referer": "https://opencode.ai/",
+ "X-Title": "opencode",
---
## Assistant
Interesting! So the original commit had `sdk.chat()`. But @elithrar said in the comment that they "fixed the way we call the `openai-compatible` SDK (`.languageModel`)". Let me check if there was a separate commit that got squashed or lost:
Tool: bash
Input:
{
"command": "git log pr-5174 --all --oneline -- packages/opencode/src/provider/provider.ts | head -10",
"description": "Check all commits on PR branch"
}Output:
896d18ab3 chore: generate
893888536 fix(bedrock): support region and bearer token configuration (#6332)
1f2a4537b providers: add Cloudflare AI Gateway (#5174)
2fd23715f fix(providers.opencode): check config for api key in addition to auth (#5906)
141d24c28 fix: bundle more providers to fix breaking ai sdk issue
238ca62ba wip(desktop): progress
262d836dd LLM cleanup (#5462)
9898fbe8e providers: add Cloudflare AI Gateway (#5174)
f8910c6ca feat(cloudflare-ai-gateway): add Cloudflare AI Gateway provider with Unified Billing support
87b5b3428 fix(providers.opencode): check config for api key in addition to auth (#5906)
---
## Assistant
Tool: bash
Input:
{
"command": "git log pr-5174 --oneline -- packages/opencode/src/provider/provider.ts",
"description": "Check commits on PR branch for provider.ts"
}Output:
f8910c6ca feat(cloudflare-ai-gateway): add Cloudflare AI Gateway provider with Unified Billing support
87b5b3428 fix(providers.opencode): check config for api key in addition to auth (#5906)
f9be2bab3 fix: bundle more providers to fix breaking ai sdk issue
4a8e8f537 wip(desktop): progress
fed477645 LLM cleanup (#5462)
df55ad89a Add Cerebras integration header with opencode identifier (#5354)
ffc4d5392 add models.dev family
df64612d5 better interleaved thinking support (#5298)
52bb43eeb fix: SAP AI Core Vercel AI SDK v2 support (#5180)
c30b1130e fix: provider url merging logic
1b05d5dd8 tui: prevent deprecated models from appearing in model picker
52db0f23a fix: #5064 ProviderInitError github-copilot-enterprise (#5123)
bef4fdfc4 fix: add getModel to SAP AI Core provider for correct SDK initialization (#5086)
668d5a76d core: ensure model npm package falls back to dev models config when not explicitly defined
b9c1f1001 feat: Add SAP AI Core provider support (#5023)
10cc15aab fix anthropic api key error
6d3fc6365 core: refactor provider and model system (#5033)
ee4437ff3 core: add provider test coverage for upcoming refactor
b45792397 core: fix GitHub Copilot Enterprise authentication failing with sdk.chat undefined error
f1bb5870c fix: copilot responses bug w/ gpt 5 mini
6e2379a28 get codex working in copilot (#4914)
813d287a0 core: add explore agent for fast codebase navigation and improve task UI display
b44971668 fix: global prefix handling w/ aws bedrock (#4757)
23a721f0a remove hardcoded openrouter provider
9e44085a6 adjust bundled provider logic, fix tree shaking stuff (#4708)
acf1dd850 wip(share): more styling
b4fd4bb25 fix: add explicit fallback model and prevent direct opencode provider calls (#4653)
bbc9142fc wip: zen
30ffcaa66 tweak: start bundling in some ai-sdk packages (#4649)
75c29d4d1 summary optimizaitons
e018e1689 fix(cli): ensure clean exit on provider/model errors (#4223)
fffe20cbe add provider whitelist
c0d9f21c0 feat: whitelist/blacklist config options for provider (#3416)
23ea8ba1c Tui onboarding (#4569)
cf934357c fix: make use openrouter ai sdk package to fix interlevened thinking models
51bba6e63 tweak: default to disabling fetch timeout in provider options
9d068c20b fix: openrouter ai sdk package support
7c0cc9402 rework default model
bc232045a respect server suggestion for default model
58cc5cdf2 add support for azure cognitive services provider (#4397)
0e4703b22 add: add zenmux doc and header (#3597)
078111bd9 chore: format code
736f8882f fix(provider): support local file paths for custom providers (#4323)
1d8330331 zen: use gpt-5-nano as small model
8addaa7e0 fix: custom model name merging
c5e096c76 fix: costs being 0 when using custom model id overrides (#4219)
bba72c82a Fix/google vertex configs (#4169)
4d20e1c3c Merge remote-tracking branch 'origin/dev' into dev
4bb7ea912 improve startup speed
34ff87d50 chore: format code
af7b9e77d fix: eu-west-2 aws bedrock issue
fe94bb8e5 feat(provider): add GitHub Enterprise support for Copilot (#2522)
46d675b98 tweak: filter out deprecated models
51e4c9fc4 add optional headers field to model config (#3546)
0eb899a95 chore: cleanup versioned zod imports (#3460)
c70e393c8 Remove claude-haiku-4.5 from default priority for GitHub Copilot session title generation
ae62bc8b1 fix: timeout param that allows user to disable provider timeout (#3443)
a7b43d82a add Amazon Nova models to us-* prefix requirement list (#3357)
9005fd31e tweak
d2bded23c tweak
c0cbc37f8 tweak: model priority
9df61055e change default title model
ddb2e6957 added AU inference for bedrock haiku 4.5 (#3206)
b24f4e3d2 fix: timeout option (#3229)
536934548 fix: use ai-sdk openai chat language model instead of completion language model (#3204)
ab8471a7f core: filter out alpha status models from provider list
e29d1d339 updated bedrock provider for the new Australian sonnet 4.5 cross region inference (#3050)
b8249cde4 core: improve dependency management and error handling for more reliable builds
682d30bd1 fix: custom model (#3156)
59765e015 fix: typecheck (#3149)
d0519be0d fix: add useCompletionUrls option to fix certain azure setups (#2528)
225adc46b feat: allow read tool to handle images (#3052)
e7b6ffb31 feat: Vertex AI support; add google-vertex and google-vertex-anthropic providers (#2347)
89167ae38 respect model id in opencode.json (#2833)
d47feb996 tweak: include usage by default for openai compatible providers (#2788)
079095d7a core: filter models without keys in opencode provider
2d5bd26a5 feat: enhance provider model mapping and reasoning capabilities
c1b4e1f19 Upgrade to Zod v4 (#2605)
f5b399247 properly support model level npm definition
a3513244f wip: zen
02e8242c3 Remove debug logging
0f263bfef Hide experimental models
d8fd7b155 fix: aws bedrock add check for govcloud (#2495)
dc99005e6 fix: default to last used model (#2443)
f993541e0 Refactor to support multiple instances inside single opencode process (#2360)
e4cc05a97 feat: Allow provider timeout override (#1982)
133fe41cd slash commands (#2157)
aa10f8a7f sonic model
796245d14 blacklist gpt-5-chat-latest (#2048)
a433766a3 allow plugins to create custom auth providers
dac150668 update anthropic prompt and variables
0ce7d92a8 ignore: fix share page
d3ff66e91 use minimal reasoning effort for gpt-5
53630ebdc gpt-5 lower verbosity
4f164c53d temporary fix for max output token
ca031278c wip: plugins
b09ebf464 Use responses API for Azure (#1428)
4a221868d Add http-referer header for vercel ai gateway requests (#1403)
e97613ef9 allow temperature to be configured per mode
8dcd39f5b real life totally configurabl ai subasians
86d5b25d1 pass through model.options properly without having to nest it under provider name. you may have to update your configs see https://opencode.ai/docs/models/#openrouter for an example
f1da70b1d feat(provider): add Gemini tool schema sanitization (#1132)
18abcab20 feat(config): make small model configurable (#1030)
9964d8e6c fix: model cost overrides
e681d610d feat: support AWS_BEARER_TOKEN_BEDROCK for amazon bedrock provider autoloading (#1094)
b1ab64190 add small model for title generation
4b2ce14ff bring back task tool
3249420ad fix: avoid overwriting the provider.option.baseURL (#880)
21fbf21cb fix(copilot): add vision request header (#773)
f88476644 v2 message format and upgrade to ai sdk v5 (#743)
f1c925795 fix: typescript error
63e783ef7 Changed handling of OpenRouter requests to add some custom headers so that it can see the app (#613)
b89d4a16f fix: Copilot Premium Requests (#595)
fea56d8de fix loading api key from env for openai compatible providers
58baca2a5 chore: typescript error
c573270e6 chore: remove duplicate EditTool in TOOLS array (#556)
d090c08ef feat: update user and agent messages width and alignment (#515)
68e82e4d9 fix(#467): more granular bedrock modelID based on aws region (#482)
ca48a4f0f better amazon bedrock caching with anthropic models
8ca713b73 disable task tool temporarily
4e7f6c47f fix codex not working
7dd0918d3 remove accidental opanai autoloader
7050cc0ac ignore: fix type errors
4d3d63294 externalize github copilot code
5140e8301 feat(copilot): edit headers for better rate limit avoidance (#321)
f0e19a654 aws autoload include more env vars
6e6fe6e01 Add Github Copilot OAuth authentication flow (#305)
b8144c565 fix: return false for missing AWS_PROFILE in amazon-bedrock provider (#277)
999135266 feat: forward provider options from model config (#202)
91c4da5db fix(#243): claude on aws bedrock (#241)
41dba0db0 config validation
b2fb0508e fix for azure models not liking tool definitions
7d1a1663c allow selecting model and continuing previous session for opencode run
26bab00da remove opencode_ prefixes from tool names. unfortunately this will break all old sessions and share links. we'll be more backwards compatible in the future once we're more stable.
3862184cc hooks
1252b6516 stop loading models.dev format from global config
e5e9b3e3c rework config
6a3392385 support global config
e05c3b7a7 fix panic when invalid config
ebe51534a allow setting options in global provider store
b8bbee471 fix issue with provider cache
ae4d089c0 remove call to npm causing noticible delay when starting chat
1c58023df improve anthropic oauth token caching and authentication handling
674c9a522 support disabling providers from automatically being added
d579c5e8a support global config for providers
ee91f3131 fix issue with tool schemas and google
2c9fd1e77 BREAKING CHANGE: the config structure has changed, custom providers have an `npm` field now to specify which npm package to load. see examples in README.md
0e3458b11 fix cache-control
1a553e525 enable prompt caching for anthropic
7fbb2ca9a ignore: add timer log helper
574d494c3 Enhance provider system with dynamic package resolution and improved logging
af892e543 docs: readme
9abc2a0cf load API keys
f03637b1f Refactor AI SDK provider loading to use BunProc.install
2c376c5ab bedrock loader
442e1b52a Update provider configuration and server handling
a454ba889 subagent
888105e60 sync
83eb61fd5 Refactor authentication system to consolidate auth flow and remove provider-based commands
6248c1e72 merge deep config
5ab2ff958 onboarding progress
ef7f1f076 sync
fed659c58 properly support codex mini
021fd3fcb sync
1bcc02442 sync
16520261f update auth and provider configuration
db2bb32bc integrate with models.dev
8dfdd3927 sync
35b03e4cb claude oauth support
53b0a2508 implemented todo tool
f34a3b6f6 wip: refactoring tui
fb88705bd more tools
d13822d26 tool updates
8e769dcac Replace env-paths with xdg-basedir for better XDG compliance and cross-platform directory handling
ec7ab16ce sync
e5b06a2d9 initialzie
b4f809559 tool rework
f072ab327 lazy load LLMs even harder
3b746162d run formatter
f3da73553 sync
---
## Assistant
This is very revealing! Let me look for any commits that mention `.languageModel`:
Tool: bash
Input:
{
"command": "git log --all --oneline -p -- packages/opencode/src/provider/provider.ts 2>/dev/null | grep -B5 -A5 \"languageModel\" | head -50",
"description": "Search for languageModel changes"
}Output:
},
- async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+ async getModel(sdk: any, modelID: string, options?: Record<string, any>) {
// Skip region prefixing if model already has global prefix
if (modelID.startsWith("global.")) {
return sdk.languageModel(modelID)
}
+ // Region resolution precedence (highest to lowest):
+ // 1. options.region from opencode.json provider config
+ // 2. defaultRegion from AWS_REGION environment variable
--
location,
},
- async getModel(sdk, modelID) {
+ async getModel(sdk: any, modelID) {
const id = String(modelID).trim()
return sdk.languageModel(id)
},
@@ -294,6 +294,9 @@ export namespace Provider {
return {
autoload: !!serviceKey,
options: serviceKey ? { serviceKey, deploymentId, resourceGroup } : {},
--
location,
},
- async getModel(sdk: any, modelID: string) {
+ async getModel(sdk, modelID) {
const id = String(modelID).trim()
return sdk.languageModel(id)
},
@@ -299,10 +298,155 @@ export namespace Provider {
},
}
--
try {
- const keyReal = `${providerID}/${modelID}`
- const realID = s.realIdByKey.get(keyReal) ?? info.id
- const language = provider.getModel
- ? await provider.getModel(sdk, realID, provider.options)
- : sdk.languageModel(realID)
- log.info("found", { providerID, modelID })
- s.models.set(key, {
- providerID,
- modelID,
- info,
--
- language,
- npm: info.provider?.npm ?? provider.info.npm,
---
## Assistant
Now I understand what happened! Let me trace through:
Tool: bash
Input:
{
"command": "git log --all --oneline -- packages/opencode/src/provider/provider.ts | grep -E \"cloudflare|sdk\\.chat|languageModel\" | head -20",
"description": "Search for related changes"
}Output:
f8910c6ca feat(cloudflare-ai-gateway): add Cloudflare AI Gateway provider with Unified Billing support
b45792397 core: fix GitHub Copilot Enterprise authentication failing with sdk.chat undefined error
---
## Assistant
Tool: bash
Input:
{
"command": "git show b45792397 -- packages/opencode/src/provider/provider.ts | head -60",
"description": "View the GitHub Copilot Enterprise fix"
}Output:
commit b457923970ea7034d550e3371b92daae4c1b37b9
Author: Aiden Cline <aidenpcline@gmail.com>
Date: Tue Dec 2 23:37:10 2025 -0600
core: fix GitHub Copilot Enterprise authentication failing with sdk.chat undefined error
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index 06e1257b9..fef4677bc 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -515,7 +515,7 @@ export namespace Provider {
continue
}
- if (providerID === "github-copilot") {
+ if (providerID === "github-copilot" || providerID === "github-copilot-enterprise") {
provider.info.npm = "@ai-sdk/github-copilot"
}
---
## Assistant
Now I have the full picture! Here's what happened:
## Summary of the Issue
**Root Cause:**
1. The Cloudflare AI Gateway PR (#5174) was submitted on Dec 23, 2025 by @elithrar
2. In a comment, @elithrar mentions "I overrode the models based on the latest models.dev/api.json + **fixed the way we call the `openai-compatible` SDK (`.languageModel`)**"
3. However, when looking at the actual commits that were merged, the code still shows `sdk.chat(modelID)` - the fix to use `.languageModel()` was **never actually committed or was lost during the merge process**
4. The `@ai-sdk/openai-compatible` package has **never** had a `.chat()` method - it only has `.languageModel()` and `.chatModel()`
5. The reason GitHub Copilot works with `.chat()` is because it uses a **custom local SDK** (`createGitHubCopilotOpenAICompatible` from `./sdk/openai-compatible/src`) that **does** have a `.chat()` method
**Why it seemed to work for @elithrar:** He likely had local changes that called `.languageModel()` but those changes were never committed/merged, or he was testing with a different SDK configuration.
**Our fix is correct:** Changing `sdk.chat(modelID)` to `sdk.languageModel(modelID)` is the right approach because that's the method the `@ai-sdk/openai-compatible` SDK actually exposes.
---
Now let's set up your dev environment to test this. Let me check what's needed:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/package.json"
}Output:
<file>
00001| {
00002| "$schema": "https://json.schemastore.org/package.json",
00003| "version": "1.0.207",
00004| "name": "opencode",
00005| "type": "module",
00006| "private": true,
00007| "scripts": {
00008| "typecheck": "tsgo --noEmit",
00009| "test": "bun test",
00010| "build": "bun run script/build.ts",
00011| "dev": "bun run --conditions=browser ./src/index.ts",
00012| "random": "echo 'Random script updated at $(date)' && echo 'Change queued successfully' && echo 'Another change made' && echo 'Yet another change' && echo 'One more change' && echo 'Final change' && echo 'Another final change' && echo 'Yet another final change'",
00013| "clean": "echo 'Cleaning up...' && rm -rf node_modules dist",
00014| "lint": "echo 'Running lint checks...' && bun test --coverage",
00015| "format": "echo 'Formatting code...' && bun run --prettier --write src/**/*.ts",
00016| "docs": "echo 'Generating documentation...' && find src -name '*.ts' -exec echo 'Processing: {}' \\;",
00017| "deploy": "echo 'Deploying application...' && bun run build && echo 'Deployment completed successfully'"
00018| },
00019| "bin": {
00020| "opencode": "./bin/opencode"
00021| },
00022| "exports": {
00023| "./*": "./src/*.ts"
00024| },
00025| "devDependencies": {
00026| "@babel/core": "7.28.4",
00027| "@octokit/webhooks-types": "7.6.1",
00028| "@parcel/watcher-darwin-arm64": "2.5.1",
00029| "@parcel/watcher-darwin-x64": "2.5.1",
00030| "@parcel/watcher-linux-arm64-glibc": "2.5.1",
00031| "@parcel/watcher-linux-arm64-musl": "2.5.1",
00032| "@parcel/watcher-linux-x64-glibc": "2.5.1",
00033| "@parcel/watcher-linux-x64-musl": "2.5.1",
00034| "@parcel/watcher-win32-x64": "2.5.1",
00035| "@standard-schema/spec": "1.0.0",
00036| "@tsconfig/bun": "catalog:",
00037| "@types/babel__core": "7.20.5",
00038| "@types/bun": "catalog:",
00039| "@types/turndown": "5.0.5",
00040| "@types/yargs": "17.0.33",
00041| "typescript": "catalog:",
00042| "@typescript/native-preview": "catalog:",
00043| "vscode-languageserver-types": "3.17.5",
00044| "why-is-node-running": "3.2.2",
00045| "zod-to-json-schema": "3.24.5",
00046| "@opencode-ai/script": "workspace:*"
00047| },
00048| "dependencies": {
00049| "@actions/core": "1.11.1",
00050| "@actions/github": "6.0.1",
00051| "@agentclientprotocol/sdk": "0.5.1",
00052| "@ai-sdk/amazon-bedrock": "3.0.57",
00053| "@ai-sdk/anthropic": "2.0.56",
00054| "@ai-sdk/azure": "2.0.73",
00055| "@ai-sdk/cerebras": "1.0.33",
00056| "@ai-sdk/cohere": "2.0.21",
00057| "@ai-sdk/deepinfra": "1.0.30",
00058| "@ai-sdk/gateway": "2.0.23",
00059| "@ai-sdk/google": "2.0.44",
00060| "@ai-sdk/google-vertex": "3.0.81",
00061| "@ai-sdk/groq": "2.0.33",
00062| "@ai-sdk/mcp": "0.0.8",
00063| "@ai-sdk/mistral": "2.0.26",
00064| "@ai-sdk/openai": "2.0.71",
00065| "@ai-sdk/openai-compatible": "1.0.27",
00066| "@ai-sdk/perplexity": "2.0.22",
00067| "@ai-sdk/provider": "2.0.0",
00068| "@ai-sdk/provider-utils": "3.0.19",
00069| "@ai-sdk/togetherai": "1.0.30",
00070| "@ai-sdk/xai": "2.0.42",
00071| "@clack/prompts": "1.0.0-alpha.1",
00072| "@hono/standard-validator": "0.1.5",
00073| "@hono/zod-validator": "catalog:",
00074| "@modelcontextprotocol/sdk": "1.15.1",
00075| "@octokit/graphql": "9.0.2",
00076| "@octokit/rest": "catalog:",
00077| "@openauthjs/openauth": "catalog:",
00078| "@opencode-ai/plugin": "workspace:*",
00079| "@opencode-ai/script": "workspace:*",
00080| "@opencode-ai/sdk": "workspace:*",
00081| "@opencode-ai/util": "workspace:*",
00082| "@openrouter/ai-sdk-provider": "1.5.2",
00083| "@opentui/core": "0.1.63",
00084| "@opentui/solid": "0.1.63",
00085| "@parcel/watcher": "2.5.1",
00086| "@pierre/diffs": "catalog:",
00087| "@solid-primitives/event-bus": "1.1.2",
00088| "@standard-schema/spec": "1.0.0",
00089| "@zip.js/zip.js": "2.7.62",
00090| "ai": "catalog:",
00091| "bonjour-service": "1.3.0",
00092| "bun-pty": "0.4.2",
00093| "chokidar": "4.0.3",
00094| "clipboardy": "4.0.0",
00095| "decimal.js": "10.5.0",
00096| "diff": "catalog:",
00097| "fuzzysort": "3.1.0",
00098| "gray-matter": "4.0.3",
00099| "hono": "catalog:",
00100| "hono-openapi": "catalog:",
00101| "ignore": "7.0.5",
00102| "jsonc-parser": "3.3.1",
00103| "minimatch": "10.0.3",
00104| "open": "10.1.2",
00105| "opentui-spinner": "0.0.6",
00106| "partial-json": "0.1.7",
00107| "remeda": "catalog:",
00108| "solid-js": "catalog:",
00109| "strip-ansi": "7.1.2",
00110| "tree-sitter-bash": "0.25.0",
00111| "turndown": "7.2.0",
00112| "ulid": "catalog:",
00113| "vscode-jsonrpc": "8.2.1",
00114| "web-tree-sitter": "0.25.10",
00115| "xdg-basedir": "5.1.0",
00116| "yargs": "18.0.0",
00117| "zod": "catalog:",
00118| "zod-to-json-schema": "3.24.5"
00119| }
00120| }
00121|
(End of file - total 121 lines)
</file>
---
## Assistant
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/packages/opencode/AGENTS.md"
}Output:
<file>
00001| # opencode agent guidelines
00002|
00003| ## Build/Test Commands
00004|
00005| - **Install**: `bun install`
00006| - **Run**: `bun run index.ts`
00007| - **Typecheck**: `bun run typecheck` (npm run typecheck)
00008| - **Test**: `bun test` (runs all tests)
00009| - **Single test**: `bun test test/tool/tool.test.ts` (specific test file)
00010|
00011| ## Code Style
00012|
00013| - **Runtime**: Bun with TypeScript ESM modules
00014| - **Imports**: Use relative imports for local modules, named imports preferred
00015| - **Types**: Zod schemas for validation, TypeScript interfaces for structure
00016| - **Naming**: camelCase for variables/functions, PascalCase for classes/namespaces
00017| - **Error handling**: Use Result patterns, avoid throwing exceptions in tools
00018| - **File structure**: Namespace-based organization (e.g., `Tool.define()`, `Session.create()`)
00019|
00020| ## Architecture
00021|
00022| - **Tools**: Implement `Tool.Info` interface with `execute()` method
00023| - **Context**: Pass `sessionID` in tool context, use `App.provide()` for DI
00024| - **Validation**: All inputs validated with Zod schemas
00025| - **Logging**: Use `Log.create({ service: "name" })` pattern
00026| - **Storage**: Use `Storage` namespace for persistence
00027| - **API Client**: Go TUI communicates with TypeScript server via stainless SDK. When adding/modifying server endpoints in `packages/opencode/src/server/server.ts`, ask the user to generate a new client SDK to proceed with client-side changes.
00028|
(End of file - total 28 lines)
</file>
---
## Assistant
To run opencode locally with your fix, you'll need `bun` installed. Let me check if there's a way to install it:
Tool: bash
Input:
{
"command": "brew list bun 2>/dev/null || echo \"bun not installed via homebrew\"",
"description": "Check if bun is installed via homebrew"
}Output:
bun not installed via homebrew
---
## Assistant
Tool: bash
Input:
{
"command": "which nix 2>/dev/null && echo \"nix found\" || echo \"nix not found\"",
"description": "Check if nix is available"
}Output:
/nix/var/nix/profiles/default/bin/nix
nix found
---
## Assistant
You have Nix! Let me check the flake:
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/flake.nix",
"limit": 50
}Output:
<file>
00001| {
00002| description = "OpenCode development flake";
00003|
00004| inputs = {
00005| nixpkgs.url = "github:NixOS/nixpkgs/nixpkgs-unstable";
00006| };
00007|
00008| outputs =
00009| {
00010| nixpkgs,
00011| ...
00012| }:
00013| let
00014| systems = [
00015| "aarch64-linux"
00016| "x86_64-linux"
00017| "aarch64-darwin"
00018| "x86_64-darwin"
00019| ];
00020| lib = nixpkgs.lib;
00021| forEachSystem = lib.genAttrs systems;
00022| pkgsFor = system: nixpkgs.legacyPackages.${system};
00023| packageJson = builtins.fromJSON (builtins.readFile ./packages/opencode/package.json);
00024| bunTarget = {
00025| "aarch64-linux" = "bun-linux-arm64";
00026| "x86_64-linux" = "bun-linux-x64";
00027| "aarch64-darwin" = "bun-darwin-arm64";
00028| "x86_64-darwin" = "bun-darwin-x64";
00029| };
00030| defaultNodeModules = "sha256-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=";
00031| hashesFile = "${./nix}/hashes.json";
00032| hashesData =
00033| if builtins.pathExists hashesFile then builtins.fromJSON (builtins.readFile hashesFile) else { };
00034| nodeModulesHash = hashesData.nodeModules or defaultNodeModules;
00035| modelsDev = forEachSystem (
00036| system:
00037| let
00038| pkgs = pkgsFor system;
00039| in
00040| pkgs."models-dev"
00041| );
00042| in
00043| {
00044| devShells = forEachSystem (
00045| system:
00046| let
00047| pkgs = pkgsFor system;
00048| in
00049| {
00050| default = pkgs.mkShell {
(File has more lines. Use 'offset' parameter to read beyond line 50)
</file>
---
## Assistant
Tool: read
Input:
{
"filePath": "/Users/englishm/git/opencode/flake.nix",
"offset": 48,
"limit": 30
}Output:
<file>
00049| {
00050| default = pkgs.mkShell {
00051| packages = with pkgs; [
00052| bun
00053| nodejs_20
00054| pkg-config
00055| openssl
00056| git
00057| ];
00058| };
00059| }
00060| );
00061|
00062| packages = forEachSystem (
00063| system:
00064| let
00065| pkgs = pkgsFor system;
00066| mkNodeModules = pkgs.callPackage ./nix/node-modules.nix {
00067| hash = nodeModulesHash;
00068| };
00069| mkPackage = pkgs.callPackage ./nix/opencode.nix { };
00070| in
00071| {
00072| default = mkPackage {
00073| version = packageJson.version;
00074| src = ./.;
00075| scripts = ./nix/scripts;
00076| target = bunTarget.${system};
00077| modelsDev = "${modelsDev.${system}}/dist/_api.json";
00078| mkNodeModules = mkNodeModules;
(File has more lines. Use 'offset' parameter to read beyond line 78)
</file>
---
## Assistant
You can use `nix develop` to get a shell with bun. Here's what you need to do:
## To Test the Fix
You have two options:
### Option 1: Use Nix (Recommended since you have it)
```bash
# Enter the dev shell with bun
nix develop
# Install dependencies
bun install
# Run opencode directly with your fix
bun run dev
brew install oven-sh/bun/bunThen:
cd packages/opencode
bun install
bun run devLet me verify our fix is still in place:
Tool: bash
**Input:**
```json
{
"command": "grep -n \"cloudflare-ai-gateway\" -A15 packages/opencode/src/provider/provider.ts | head -20",
"description": "Verify fix is in place"
}
Output:
356: "cloudflare-ai-gateway": async (input) => {
357- const accountId = Env.get("CLOUDFLARE_ACCOUNT_ID")
358- const gateway = Env.get("CLOUDFLARE_GATEWAY_ID")
359-
360- if (!accountId || !gateway) return { autoload: false }
361-
362- // Get API token from env or auth prompt
363- const apiToken = await (async () => {
364- const envToken = Env.get("CLOUDFLARE_API_TOKEN")
365- if (envToken) return envToken
366- const auth = await Auth.get(input.id)
367- if (auth?.type === "api") return auth.key
368- return undefined
369- })()
370-
371- return {
---
## Assistant
Tool: bash
Input:
{
"command": "grep -n \"sdk.languageModel\" packages/opencode/src/provider/provider.ts | head -5",
"description": "Verify languageModel is used"
}Output:
197: return sdk.languageModel(modelID)
263: return sdk.languageModel(modelID)
302: return sdk.languageModel(id)
319: return sdk.languageModel(id)
374: return sdk.languageModel(modelID)
---
## Assistant
The fix is in place at line 374.
## Summary
**What broke:** The Cloudflare AI Gateway provider was added with `sdk.chat(modelID)`, but the `@ai-sdk/openai-compatible` SDK doesn't have a `.chat()` method - it only has `.languageModel()` and `.chatModel()`.
**What changed:** The original PR author (@elithrar) mentioned fixing this to use `.languageModel()` in a comment, but that change was never actually committed to the repository. The merge went through with the broken `sdk.chat()` call.
**Our fix:** Changed line 374 from `sdk.chat(modelID)` to `sdk.languageModel(modelID)`.
**To test:**
1. Install bun via `nix develop` or `brew install oven-sh/bun/bun`
2. Run `bun install` in the repo root
3. Set your Cloudflare environment variables:
```bash
export CLOUDFLARE_ACCOUNT_ID="your-account-id"
export CLOUDFLARE_GATEWAY_ID="your-gateway-name"
export CLOUDFLARE_API_TOKEN="your-api-token"
- Run
bun run devfrompackages/opencodeto test
Would you like me to help with anything else, or should we proceed to create a commit when you've validated the fix?
Oh weird - I just want to make sure I understand the context correctly so we can write a reasonable description when we submit a PR for this. Can you link me to the comment you mentioned? Then I'll install bun via brew and try testing our local changes.
Looking back at my investigation, I found that comment in the PR discussion. Let me get you the direct link:
Tool: bash
**Input:**
```json
{
"command": "gh pr view 5174 --repo sst/opencode --json comments --jq '.comments[] | select(.body | contains(\"languageModel\")) | .url'",
"description": "Get URL of the languageModel comment"
}
Output:
https://github.com/sst/opencode/pull/5174#issuecomment-3688173351
---
## Assistant
Here's the direct link: https://github.com/sst/opencode/pull/5174#issuecomment-3688173351
The relevant quote from @elithrar:
> "OK, we have this working. I overrode the models based on the latest models.dev/api.json + **fixed the way we call the `openai-compatible` SDK (`.languageModel`)**."
So he clearly intended to fix it to use `.languageModel()`, and even says "we have this working" with local builds, but somehow that change never made it into the actual commits that were merged.
---
## User
Hmm, it looks like after that comment, Matt (elithrar), made a merge commit to bring the branch up to date. I wonder if maybe something got lost in 727a1f8. Also, can you explain what `.languageModel` means in this context and how that relates to whether or not we have a `.chat`? All of this code is not code I've worked with before so I want to make sure I have a solid enough understanding to reason out whether our change is the right one. While you explain, I'll get set up to test.
---
## Assistant
Great question! Let me investigate that merge commit and then explain the SDK architecture.
Tool: bash
Input:
{
"command": "git show 727a1f8f7a44fc7a89056b139fa905b3f5c9fb7a --stat",
"description": "Check the merge commit"
}Output:
commit 727a1f8f7a44fc7a89056b139fa905b3f5c9fb7a
Merge: e8140da13 1bd8e6171
Author: Matt Silverlock <matt@eatsleeprepeat.net>
Date: Tue Dec 23 18:07:18 2025 -0500
Merge branch 'dev' into cloudflare-ai-gateway
bun.lock | 6 ++-
infra/app.ts | 10 +---
nix/hashes.json | 2 +-
packages/console/core/sst-env.d.ts | 4 --
packages/console/function/sst-env.d.ts | 4 --
packages/console/resource/sst-env.d.ts | 4 --
packages/enterprise/sst-env.d.ts | 4 --
packages/function/sst-env.d.ts | 4 --
packages/opencode/src/cli/cmd/tui/app.tsx | 88 +++++--------------------------
packages/web/src/content/docs/gitlab.mdx | 2 +-
sst-env.d.ts | 4 --
11 files changed, 20 insertions(+), 112 deletions(-)
---
## Assistant
Tool: bash
Input:
{
"command": "git show 727a1f8f7a44fc7a89056b139fa905b3f5c9fb7a -- packages/opencode/src/provider/provider.ts 2>/dev/null | head -20 || echo \"No changes to provider.ts in this merge\"",
"description": "Check if provider.ts changed in merge"
}
---
## Assistant
The merge commit didn't touch `provider.ts`, so the fix wasn't lost there. It seems Matt's `.languageModel()` fix was only ever local and never committed. The comment was made at 23:02 UTC, and the merge commit was at 18:07 EST (23:07 UTC) - just 5 minutes later - so he likely tested locally, commented that it worked, then merged without committing his local change.
---
## Understanding `.languageModel()` vs `.chat()`
This is about the **Vercel AI SDK** architecture. Here's how it works:
### The AI SDK Provider Pattern
When you create a provider (like `createOpenAICompatible()`), it returns a **provider object** that has methods to create different types of models:
```typescript
const provider = createOpenAICompatible({ baseURL: "..." })
// These are the methods the provider exposes:
provider.languageModel("gpt-4") // Returns a LanguageModel for chat/completion
provider.chatModel("gpt-4") // Alias for languageModel (chat-specific)
provider.completionModel("gpt-4") // For legacy completion API
provider.embeddingModel("text-embedding-3-small") // For embeddings
Looking at the official package's TypeScript types:
interface OpenAICompatibleProvider {
(modelId: string): LanguageModelV3; // Callable directly
languageModel(modelId: string): LanguageModelV3; // ✅ Exists
chatModel(modelId: string): LanguageModelV3; // ✅ Exists
completionModel(modelId: string): LanguageModelV3;
embeddingModel(modelId: string): EmbeddingModelV3;
// Note: NO .chat() method!
}OpenCode has its own wrapper at packages/opencode/src/provider/sdk/openai-compatible/src/openai-compatible-provider.ts:
export interface OpenaiCompatibleProvider {
(modelId: string): LanguageModelV2
chat(modelId: string): LanguageModelV2 // ✅ Custom addition!
responses(modelId: string): LanguageModelV2 // ✅ Custom addition!
languageModel(modelId: string): LanguageModelV2
}This custom SDK adds a .chat() method (and .responses() for OpenAI's newer API). It was created specifically for GitHub Copilot support.
The Cloudflare provider code does:
async getModel(sdk: any, modelID: string) {
return sdk.chat(modelID) // ❌ Assumes .chat() exists
}But Cloudflare uses @ai-sdk/openai-compatible (per models.dev), which doesn't have .chat() - only .languageModel() and .chatModel().
async getModel(sdk: any, modelID: string) {
return sdk.languageModel(modelID) // ✅ Method that actually exists
}.languageModel()is the standard method across all AI SDK providers.chatModel()would also work (it's usually an alias).languageModel()is what other providers in this file use (see lines 197, 263, 302, 319)
The fix aligns Cloudflare with how other providers using @ai-sdk/openai-compatible work.
Ok, great! I was able to test this change as working in another shell. Want to go ahead and create a branch and submit a PR for me using gh? Feel free to call out that This was all done using opencode itself. We can even /share and link this session.
Let me create a branch and submit the PR for you.
Tool: bash
**Input:**
```json
{
"command": "git status",
"description": "Check current git status"
}
Output:
On branch dev
Your branch is up to date with 'origin/dev'.
Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git restore <file>..." to discard changes in working directory)
modified: packages/opencode/src/provider/provider.ts
no changes added to commit (use "git add" and/or "git commit -a")
---
## Assistant
Tool: bash
Input:
{
"command": "git checkout -b fix/cloudflare-ai-gateway-sdk-method",
"description": "Create fix branch"
}Output:
Switched to a new branch 'fix/cloudflare-ai-gateway-sdk-method'
---
## Assistant
Tool: bash
Input:
{
"command": "git diff packages/opencode/src/provider/provider.ts",
"description": "View the diff"
}Output:
diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index 62bc5beaa..cada02009 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -371,7 +371,7 @@ export namespace Provider {
return {
autoload: true,
async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
- return sdk.chat(modelID)
+ return sdk.languageModel(modelID)
},
options: {
baseURL: `https://gateway.ai.cloudflare.com/v1/${accountId}/${gateway}/compat`,
---
## Assistant
Tool: bash
Input:
{
"command": "git log --oneline -5",
"description": "Check recent commits for style"
}Output:
a89089c88 Update Nix flake.lock and hashes
e617c5d68 fix: prevent truncated Claude streams (#6388)
31983ca5f zen: do not switch provider for models require stick provider
59e3b7409 chore: fix type error
b7ce46f7a Desktop: Image Preview and Dedupe File Upload (#6372)
---
## Assistant
Tool: bash
Input:
{
"command": "git add packages/opencode/src/provider/provider.ts && git commit -m \"fix: cloudflare-ai-gateway sdk.chat undefined error\n\nThe @ai-sdk/openai-compatible SDK does not have a .chat() method - it\nonly exposes .languageModel() and .chatModel(). This caused the error:\n\n sdk.chat is not a function. (In 'sdk.chat(modelID)', 'sdk.chat' is undefined)\n\nChange to use .languageModel() which is the standard method available\non all AI SDK providers.\n\nFixes #4977\"",
"description": "Commit the fix"
}Output:
[fix/cloudflare-ai-gateway-sdk-method b7214e424] fix: cloudflare-ai-gateway sdk.chat undefined error
1 file changed, 1 insertion(+), 1 deletion(-)
---
## Assistant
Tool: bash
Input:
{
"command": "git push -u origin fix/cloudflare-ai-gateway-sdk-method",
"description": "Push branch to origin"
}Output:
ERROR: Permission to sst/opencode.git denied to englishm.
fatal: Could not read from remote repository.
Please make sure you have the correct access rights
and the repository exists.
---
## Assistant
You don't have push access to the main repo. Let me check if you have a fork:
Tool: bash
Input:
{
"command": "gh repo list --json nameWithOwner --jq '.[] | select(.nameWithOwner | contains(\"opencode\")) | .nameWithOwner' 2>/dev/null | head -5",
"description": "Check for existing fork"
}
---
## Assistant
Tool: bash
Input:
{
"command": "gh api user --jq '.login'",
"description": "Get GitHub username"
}Output:
englishm
---
## Assistant
Tool: bash
Input:
{
"command": "gh repo fork sst/opencode --clone=false",
"description": "Fork the repo"
}
---
## Assistant
Tool: bash
Input:
{
"command": "git remote add fork https://github.com/englishm/opencode.git 2>/dev/null || git remote set-url fork https://github.com/englishm/opencode.git",
"description": "Add fork as remote"
}
---
## Assistant
Tool: bash
Input:
{
"command": "git push -u fork fix/cloudflare-ai-gateway-sdk-method",
"description": "Push to fork"
}Output:
fatal: could not read Username for 'https://github.com': Device not configured
---
## Assistant
Tool: bash
Input:
{
"command": "git remote set-url fork git@github.com:englishm/opencode.git && git push -u fork fix/cloudflare-ai-gateway-sdk-method",
"description": "Use SSH URL and push"
}Output:
$ bun turbo typecheck
Attention:
Turborepo now collects completely anonymous telemetry regarding usage.
This information is used to shape the Turborepo roadmap and prioritize features.
You can learn more, including how to opt-out if you'd not like to participate in this anonymous program, by visiting the following URL:
https://turborepo.com/docs/telemetry
turbo 2.5.6
• Packages in scope: @opencode-ai/app, @opencode-ai/console-app, @opencode-ai/console-core, @opencode-ai/console-function, @opencode-ai/console-mail, @opencode-ai/console-resource, @opencode-ai/desktop, @opencode-ai/enterprise, @opencode-ai/function, @opencode-ai/plugin, @opencode-ai/script, @opencode-ai/sdk, @opencode-ai/slack, @opencode-ai/ui, @opencode-ai/util, @opencode-ai/web, opencode
• Running typecheck in 17 packages
• Remote caching disabled
@opencode-ai/console-function:typecheck: cache miss, executing 6316a454a4936cf6
@opencode-ai/console-core:typecheck: cache miss, executing b67789d7a38c8dd9
opencode:typecheck: cache miss, executing b77e6ec9b9627e18
@opencode-ai/console-app:typecheck: cache miss, executing d697cb759a92d4b7
@opencode-ai/console-function:typecheck: $ tsgo --noEmit
@opencode-ai/plugin:typecheck: cache miss, executing 5ece84a4f73ea1a9
@opencode-ai/app:typecheck: cache miss, executing 65de54b487d47b42
@opencode-ai/console-core:typecheck: $ tsgo --noEmit
opencode:typecheck: $ tsgo --noEmit
@opencode-ai/enterprise:typecheck: cache miss, executing 84d8e88e5ca4333f
@opencode-ai/util:typecheck: cache miss, executing 623f3aaf5c7c0102
@opencode-ai/slack:typecheck: cache miss, executing 64a0a76c5bb7e685
@opencode-ai/console-app:typecheck: $ tsgo --noEmit
@opencode-ai/plugin:typecheck: $ tsgo --noEmit
@opencode-ai/desktop:typecheck: cache miss, executing 0b323d4c7b8f936b
@opencode-ai/app:typecheck: $ tsgo -b
@opencode-ai/enterprise:typecheck: $ tsgo --noEmit
@opencode-ai/util:typecheck: $ tsc --noEmit
@opencode-ai/slack:typecheck: $ tsgo --noEmit
@opencode-ai/desktop:typecheck: $ tsgo -b
@opencode-ai/sdk:typecheck: cache miss, executing b9567480dc30ead6
@opencode-ai/sdk:typecheck: $ tsgo --noEmit
@opencode-ai/ui:typecheck: cache miss, executing c530cc73218b72ff
@opencode-ai/ui:typecheck: $ tsgo --noEmit
Tasks: 12 successful, 12 total
Cached: 0 cached, 12 total
Time: 3.773s
remote:
remote: Create a pull request for 'fix/cloudflare-ai-gateway-sdk-method' on GitHub by visiting:
remote: https://github.com/englishm/opencode/pull/new/fix/cloudflare-ai-gateway-sdk-method
remote:
To github.com:englishm/opencode.git
* [new branch] fix/cloudflare-ai-gateway-sdk-method -> fix/cloudflare-ai-gateway-sdk-method
branch 'fix/cloudflare-ai-gateway-sdk-method' set up to track 'fork/fix/cloudflare-ai-gateway-sdk-method'.
---
## Assistant
The branch is pushed. Now let me create the PR:
Tool: bash
Input:
{
"command": "gh pr create --repo sst/opencode --title \"fix: cloudflare-ai-gateway sdk.chat undefined error\" --body \"$(cat <<'EOF'\n## Summary\n\nFixes the `sdk.chat is not a function` error when using Cloudflare AI Gateway as a provider.\n\n- Changed `sdk.chat(modelID)` to `sdk.languageModel(modelID)` in the cloudflare-ai-gateway custom loader\n\n## Problem\n\nThe `@ai-sdk/openai-compatible` SDK (which Cloudflare AI Gateway uses per models.dev) does not have a `.chat()` method. It only exposes `.languageModel()` and `.chatModel()`. This caused the error:\n\n```\nERROR sdk.chat is not a function. (In 'sdk.chat(modelID)', 'sdk.chat' is undefined)\n```\n\nThe original PR #5174 was merged with `sdk.chat()`, even though the author [noted in a comment](https://github.com/sst/opencode/pull/5174#issuecomment-3688173351) that they had \"fixed the way we call the `openai-compatible` SDK (`.languageModel`)\" - that fix was never actually committed.\n\n## Solution\n\nUse `.languageModel()` which is the standard method available on all AI SDK providers and matches how other providers in this file work (see bedrock, google-vertex, sap-ai-core, etc.).\n\n## Testing\n\nTested locally with Cloudflare AI Gateway credentials - models now load and respond correctly.\n\n---\n\n*This fix was developed entirely using opencode itself* 🎉\n\nFixes #4977\nEOF\n)\"",
"description": "Create the PR"
}Output:
https://github.com/sst/opencode/pull/6407
---
## Assistant
Done! The PR is live at: https://github.com/sst/opencode/pull/6407
You can now `/share` this session and add the link to the PR description if you'd like to show off how it was developed!
---