Models

The models the agent runs on, made from model endpoints: an Effect AI client for the endpoint’s protocol, at its URL, with its key and headers, and the endpoint’s model on that client, set to the endpoint’s effort and output limit.

The agent makes them with LanguageModels, which layer provides over the host’s HttpClient, that the requests go out through. A test provides models of its own in its place.

Since v0.1.0

Layers

layer

Provides LanguageModels over the HttpClient the requests go out through, such as FetchHttpClient.layer.

Example (Generating text on OpenRouter with the model of an endpoint)

import { Models } from "@amends-dev/harness-amends"
import { Config, Effect, Layer } from "effect"
import { LanguageModel } from "effect/ai"
import { FetchHttpClient } from "effect/http"

const program = Effect.gen(function* () {
  const languageModels = yield* Models.LanguageModels
  const model = yield* languageModels.make({
    provider: "openrouter",
    model: "nvidia/nemotron-3-super-120b-a12b:free",
    protocol: "openai-chat",
    url: "https://openrouter.ai/api/v1",
    apiKey: yield* Config.Redacted("OPENROUTER_API_KEY")
  })
  const response = yield* LanguageModel.generateText({ prompt: "Say hi" }).pipe(
    Effect.provideService(LanguageModel.LanguageModel, model)
  )
  return response.text
}).pipe(Effect.provide(Models.layer.pipe(Layer.provide(FetchHttpClient.layer))))

Signature

declare const layer: Layer.Layer<LanguageModels, never, HttpClient.HttpClient>

Source

Since v0.1.0

Services

LanguageModels (class)

Makes the language model an endpoint reaches. The agent takes it as a service, so that a test provides models of its own in place of the providers’ clients.

Signature

declare class LanguageModels

Source

Since v0.1.0