feat(ai): Responses routes for Fireworks and Groq

This commit is contained in:
Aiden Cline
2026-10-31 23:51:38 -05:00
Unverified
parent 01276f9061
commit 4f01403e1e
12 changed files with 308 additions and 24 deletions
+4
View File
@@ -1517,6 +1517,10 @@ const deepseek = DeepSeek.configure({ apiKey }).model("deepseek-chat")
const fireworks = Fireworks.configure({ apiKey }).model("accounts/fireworks/models/my-model")
```
Fireworks and Groq also select their Responses API with `responses(modelID)`, or through `providers/fireworks/responses`
and `providers/groq/responses`. They use the provider-neutral Open Responses protocol with `store: false`. Groq rejects
`include`, so its reasoning is not replayed as encrypted content.
The former `OpenAICompatible.baseten`, `.cerebras`, `.deepinfra`, `.deepseek`, `.fireworks`, `.groq`, and `.togetherai` presets are replaced by the top-level `Baseten`, `Cerebras`, `DeepInfra`, `DeepSeek`, `Fireworks`, `Groq`, and `TogetherAI` exports. Use `CloudflareAIGateway` and `CloudflareWorkersAI` directly; each has its own module. `OpenAICompatible` configures generic endpoints with an explicit `baseURL`.
### Provider entrypoints
+2
View File
@@ -60,6 +60,7 @@
"./providers/digitalocean/responses": "./src/providers/digitalocean/responses.ts",
"./providers/elevenlabs": "./src/providers/elevenlabs.ts",
"./providers/fireworks": "./src/providers/fireworks.ts",
"./providers/fireworks/responses": "./src/providers/fireworks/responses.ts",
"./providers/google": "./src/providers/google.ts",
"./providers/google-vertex": "./src/providers/google-vertex.ts",
"./providers/google-vertex-messages": "./src/providers/google-vertex-messages.ts",
@@ -73,6 +74,7 @@
"./providers/google/images": "./src/providers/google/images.ts",
"./providers/google/interactions": "./src/providers/google/interactions.ts",
"./providers/groq": "./src/providers/groq.ts",
"./providers/groq/responses": "./src/providers/groq/responses.ts",
"./providers/meta": "./src/providers/meta.ts",
"./providers/meta/chat": "./src/providers/meta/chat.ts",
"./providers/meta/images": "./src/providers/meta/images.ts",
+34 -12
View File
@@ -1,10 +1,12 @@
import type { ProviderPackage } from "../provider-package.js"
import { OpenAIChat } from "../protocols/openai-chat.js"
import { OpenResponses } from "../protocols/open-responses.js"
import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
import { Route, type RouteDefaultsInput } from "../route/client.js"
import { Endpoint } from "../route/endpoint.js"
import { ProviderID, type ModelID } from "../schema/index.js"
import type { OpenAIProviderOptionsInput } from "./openai-options.js"
import type { OpenResponsesProviderOptionsInput } from "./open-responses-options.js"
export const id = ProviderID.make("fireworks")
const baseURL = "https://api.fireworks.ai/inference/v1"
@@ -12,11 +14,11 @@ const baseURL = "https://api.fireworks.ai/inference/v1"
export type Config = Omit<RouteDefaultsInput, "providerOptions"> &
ProviderAuthOption<"optional"> & {
readonly baseURL?: string
readonly providerOptions?: OpenAIProviderOptionsInput
readonly providerOptions?: OpenAIProviderOptionsInput | OpenResponsesProviderOptionsInput
}
export type Settings = ProviderPackage.Settings &
OpenAIProviderOptionsInput & {
export type Settings<Options = OpenAIProviderOptionsInput> = ProviderPackage.Settings &
Options & {
readonly apiKey?: string | false
readonly baseURL?: string
}
@@ -30,35 +32,55 @@ export const route = Route.make({
framing: OpenAIChat.framing,
})
export const routes = [route]
const responsesRoute = Route.make({
id: "fireworks-responses",
provider: id,
providerMetadataKey: "fireworks",
protocol: OpenResponses.protocol,
endpoint: Endpoint.path("/responses", { baseURL }),
transport: OpenResponses.httpTransport,
defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
})
export const routes = [route, responsesRoute]
export const configure = (input: Config = {}) => {
const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
const configured = route.with({
const options = {
...defaults,
endpoint: { baseURL: endpoint ?? baseURL },
auth: AuthOptions.bearer(input, "FIREWORKS_API_KEY"),
})
}
const configured = route.with(options)
const configuredResponses = responsesRoute.with(options)
return {
id,
model: (modelID: string | ModelID) =>
configured.model<OpenAIProviderOptionsInput>({ id: modelID, compatibility: { supportsPromptCacheKey: true } }),
responses: (modelID: string | ModelID) =>
configuredResponses.model<OpenResponsesProviderOptionsInput>({ id: modelID }),
configure,
}
}
export const provider = configure()
export const responses = provider.responses
export const model: ProviderPackage.Model<Settings, OpenAIProviderOptionsInput> = (
modelID,
{ apiKey, baseURL, body, headers, ...providerOptions }: Settings = {},
) =>
configure({
export const model: ProviderPackage.Model<Settings, OpenAIProviderOptionsInput> = (modelID, settings) =>
configure(config(settings)).model(modelID)
export const responsesModel: ProviderPackage.Model<
Settings<OpenResponsesProviderOptionsInput>,
OpenResponsesProviderOptionsInput
> = (modelID, settings) => configure(config(settings)).responses(modelID)
function config({ apiKey, baseURL, body, headers, ...providerOptions }: Settings<Config["providerOptions"]> = {}) {
return {
apiKey,
baseURL,
headers,
http: body === undefined ? undefined : { body },
providerOptions,
}).model(modelID)
}
}
export * as Fireworks from "./fireworks.js"
@@ -0,0 +1,4 @@
import type { OpenResponsesProviderOptionsInput } from "../open-responses-options.js"
import type { Fireworks } from "../fireworks.js"
export { responsesModel as model } from "../fireworks.js"
export type Settings = Fireworks.Settings<OpenResponsesProviderOptionsInput>
+34 -11
View File
@@ -1,6 +1,7 @@
import { Effect, Schema } from "effect"
import type { ProviderPackage } from "../provider-package.js"
import { OpenAIChat } from "../protocols/openai-chat.js"
import { OpenResponses } from "../protocols/open-responses.js"
import { ProviderShared } from "../protocols/shared.js"
import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
import { Route, type RouteDefaultsInput } from "../route/client.js"
@@ -8,6 +9,7 @@ import { Endpoint } from "../route/endpoint.js"
import { Protocol } from "../route/protocol.js"
import { ProviderID, type ModelID, type LLMRequest } from "../schema/index.js"
import type { OpenAIProviderOptionsInput } from "./openai-options.js"
import type { OpenResponsesProviderOptionsInput } from "./open-responses-options.js"
export const id = ProviderID.make("groq")
const baseURL = "https://api.groq.com/openai/v1"
@@ -23,11 +25,11 @@ export type ProviderOptions = Pick<OpenAIProviderOptionsInput, "reasoningEffort"
export type Config = Omit<RouteDefaultsInput, "providerOptions"> &
ProviderAuthOption<"optional"> & {
readonly baseURL?: string
readonly providerOptions?: ProviderOptions
readonly providerOptions?: ProviderOptions | OpenResponsesProviderOptionsInput
}
export type Settings = ProviderPackage.Settings &
ProviderOptions & {
export type Settings<Options = ProviderOptions> = ProviderPackage.Settings &
Options & {
readonly apiKey?: string | false
readonly baseURL?: string
}
@@ -71,13 +73,26 @@ export const route = Route.make({
framing: OpenAIChat.framing,
})
// Groq's Responses API is stateless and rejects `include`, so reasoning is not replayed as encrypted content.
const responsesRoute = Route.make({
id: "groq-responses",
provider: id,
providerMetadataKey: "groq",
protocol: OpenResponses.protocol,
endpoint: Endpoint.path("/responses", { baseURL }),
transport: OpenResponses.httpTransport,
defaults: { providerOptions: { store: false } },
})
export const configure = (input: Config = {}) => {
const { apiKey: _apiKey, auth: _auth, baseURL: endpoint, ...defaults } = input
const configured = route.with({
const options = {
...defaults,
endpoint: { baseURL: endpoint ?? baseURL },
auth: AuthOptions.bearer(input, "GROQ_API_KEY"),
})
}
const configured = route.with(options)
const configuredResponses = responsesRoute.with(options)
return {
id,
model: (modelID: string | ModelID) =>
@@ -90,22 +105,30 @@ export const configure = (input: Config = {}) => {
supportsStrictMode: false,
},
}),
responses: (modelID: string | ModelID) =>
configuredResponses.model<OpenResponsesProviderOptionsInput>({ id: modelID }),
configure,
}
}
export const provider = configure()
export const responses = provider.responses
export const model: ProviderPackage.Model<Settings, ProviderOptions> = (
modelID,
{ apiKey, baseURL, body, headers, ...providerOptions }: Settings = {},
) =>
configure({
export const model: ProviderPackage.Model<Settings, ProviderOptions> = (modelID, settings) =>
configure(config(settings)).model(modelID)
export const responsesModel: ProviderPackage.Model<
Settings<OpenResponsesProviderOptionsInput>,
OpenResponsesProviderOptionsInput
> = (modelID, settings) => configure(config(settings)).responses(modelID)
function config({ apiKey, baseURL, body, headers, ...providerOptions }: Settings<Config["providerOptions"]> = {}) {
return {
apiKey,
baseURL,
headers,
http: body === undefined ? undefined : { body },
providerOptions,
}).model(modelID)
}
}
export * as Groq from "./groq.js"
@@ -0,0 +1,4 @@
import type { OpenResponsesProviderOptionsInput } from "../open-responses-options.js"
import type { Groq } from "../groq.js"
export { responsesModel as model } from "../groq.js"
export type Settings = Groq.Settings<OpenResponsesProviderOptionsInput>
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
+32
View File
@@ -104,6 +104,38 @@ describe("provider package entrypoints", () => {
}
})
test("maps Fireworks and Groq Responses entrypoints onto provider-owned routes", async () => {
const settings = {
apiKey: "fixture",
headers: { "x-test": "fixture" },
body: { future_option: true },
reasoningEffort: "low",
}
const encrypted = { store: false, include: ["reasoning.encrypted_content"] }
const selected = [
[
(await import("@opencode/ai/providers/fireworks/responses")).model("gpt-oss", settings),
"fireworks-responses",
"https://api.fireworks.ai/inference/v1",
encrypted,
],
[
(await import("@opencode/ai/providers/groq/responses")).model("gpt-oss", settings),
"groq-responses",
"https://api.groq.com/openai/v1",
{ store: false },
],
] as const
for (const [model, route, baseURL, defaults] of selected) {
expect(model.route.id).toBe(route)
expect(model.route.protocol).toBe("open-responses")
expect(model.route.endpoint).toMatchObject({ baseURL, path: "/responses" })
expect(model.route.defaults.headers).toEqual(settings.headers)
expect(model.route.defaults.http?.body).toEqual(settings.body)
expect(model.route.defaults.providerOptions).toEqual({ reasoningEffort: "low", ...defaults })
}
})
test("maps Cohere entrypoints onto native and compatibility routes", async () => {
const modules = await Promise.all([
import("@opencode/ai/providers/cohere"),
@@ -1,7 +1,7 @@
import { expect } from "bun:test"
import { Effect, Schema } from "effect"
import { ai, LLMEvent, toDefinitions } from "../../src/effect.js"
import { DigitalOcean, OpenRouter, Venice } from "../../src/providers.js"
import { DigitalOcean, Fireworks, Groq, OpenRouter, Venice } from "../../src/providers.js"
import { LLMClient } from "../../src/route.js"
import { tool } from "../../src/tool.js"
import { expectWeatherToolLoop, runWeatherToolLoop, weatherToolLoopRequest } from "../recorded-scenarios.js"
@@ -37,6 +37,20 @@ for (const item of [
env: "VENICE_API_KEY",
model: Venice.configure({ apiKey: process.env.VENICE_API_KEY ?? "fixture" }).responses("openai-gpt-54-mini"),
},
{
provider: "fireworks",
env: "FIREWORKS_API_KEY",
freeform: false,
model: Fireworks.configure({ apiKey: process.env.FIREWORKS_API_KEY ?? "fixture" }).responses(
"accounts/fireworks/models/gpt-oss-120b",
),
},
{
provider: "groq",
env: "GROQ_API_KEY",
freeform: false,
model: Groq.configure({ apiKey: process.env.GROQ_API_KEY ?? "fixture" }).responses("openai/gpt-oss-20b"),
},
]) {
const recorded = recordedTests({
prefix: `${item.provider}-responses`,
@@ -72,6 +86,8 @@ for (const item of [
}),
)
if (item.freeform === false) continue
recorded.effect.with("calls a freeform text tool", { tags: ["tool", "freeform"] }, () =>
Effect.gen(function* () {
const response = yield* LLMClient.generate(
@@ -89,3 +105,4 @@ for (const item of [
}),
)
}