@effect-uai/chat-completions

Reusable OpenAI Chat Completions provider for @effect-uai/core.

Implements the LanguageModel contract against the POST /chat/completions wire dialect (SSE streaming, messages[], tool_calls, response_format). Point it at any compatible endpoint by setting baseUrl: OpenRouter, Requesty, Groq, Together, or a self-hosted gateway.

Prefer @effect-uai/responses when the endpoint speaks the Responses protocol. Reach for chat-completions only when it does not: this dialect has no typed reasoning items and no server-side state (store / previousResponseId).

Install

pnpm add @effect-uai/chat-completions @effect-uai/core effect

ESM-only. Requires effect@>=4.0.1 and @effect-uai/core as peers.

Usage

import { Config, Effect, Layer } from "effect"
import { FetchHttpClient } from "effect/http"
import { layer as chatLayer } from "@effect-uai/chat-completions/ChatCompletions"

const provider = Layer.unwrap(
  Effect.gen(function* () {
    const apiKey = yield* Config.Redacted("LLM_API_KEY")
    return chatLayer({
      apiKey,
      baseUrl: "https://openrouter.ai/api/v1",
      provider: "openrouter",
    })
  }),
)

const layer = Layer.provide(provider, FetchHttpClient.layer)

The layer registers the generic LanguageModel tag, so provider-agnostic loop bodies run against it unchanged. model is a plain string: these gateways ship hundreds of models, so there is no typed union.

ChatConfig also accepts path, authHeader, extraHeaders, and extraBody for endpoints that diverge from the OpenRouter defaults.

Docs

https://effect-uai.com/providers/

License

MIT