Effect-TS/effect @effect/ai@0.18.0

on GitHub

Minor Changes

• #4891 0552674 Thanks @IMax153! - Make AiModel a plain Layer and remove AiPlan in favor of ExecutionPlan

  This release substantially simplifies and improves the ergonomics of using AiModel for various providers. With these changes, an AiModel now returns a plain Layer which can be used to provide services to a program that interacts with large language models.

  Before

  import { AiLanguageModel } from "@effect/ai"
  import { OpenAiClient, OpenAiLanguageModel } from "@effect/ai-openai"
  import { NodeHttpClient } from "@effect/platform-node"
  import { Config, Console, Effect, Layer } from "effect"
  
  // Produces an `AiModel<AiLanguageModel, OpenAiClient>`
  const Gpt4o = OpenAiLanguageModel.model("gpt-4o")
  
  // Generate a dad joke
  const getDadJoke = AiLanguageModel.generateText({
    prompt: "Tell me a dad joke"
  })
  
  const program = Effect.gen(function* () {
    // Build the `AiModel` into a `Provider`
    const gpt4o = yield* Gpt4o
    // Use the built `AiModel` to run the program
    const response = yield* gpt4o.use(getDadJoke)
    // Log the response
    yield* Console.log(response.text)
  })
  
  const OpenAi = OpenAiClient.layerConfig({
    apiKey: Config.redacted("OPENAI_API_KEY")
  }).pipe(Layer.provide(NodeHttpClient.layerUndici))
  
  program.pipe(Effect.provide(OpenAi), Effect.runPromise)

  After

  import { AiLanguageModel } from "@effect/ai"
  import { OpenAiClient, OpenAiLanguageModel } from "@effect/ai-openai"
  import { NodeHttpClient } from "@effect/platform-node"
  import { Config, Console, Effect, Layer } from "effect"
  
  // Produces a `Layer<AiLanguageModel, never, OpenAiClient>`
  const Gpt4o = OpenAiLanguageModel.model("gpt-4o")
  
  const program = Effect.gen(function*() {
    // Generate a dad joke
    const response = yield* AiLanguageModel.generateText({
      prompt: "Tell me a dad joke"
    })
    // Log the response
    yield* Console.log(response.text)
  ).pipe(Effect.provide(Gpt4o))
  
  const OpenAi = OpenAiClient.layerConfig({
    apiKey: Config.redacted("OPENAI_API_KEY")
  }).pipe(Layer.provide(NodeHttpClient.layerUndici))
  
  program.pipe(
    Effect.provide(OpenAi),
    Effect.runPromise
  )

  In addition, AiModel can be yield*'ed to produce a layer with no requirements.

  This shifts the requirements of building the layer into the calling effect, which is particularly useful for creating AI-powered services.

  import { AiLanguageModel } from "@effect/ai"
  import { OpenAiLanguageModel } from "@effect/ai-openai"
  import { Effect } from "effect"
  
  class DadJokes extends Effect.Service<DadJokes>()("DadJokes", {
    effect: Effect.gen(function* () {
      // Yielding the model will return a layer with no requirements
      //
      //      ┌─── Layer<AiLanguageModel>
      //      ▼
      const model = yield* OpenAiLanguageModel.model("gpt-4o")
  
      const getDadJoke = AiLanguageModel.generateText({
        prompt: "Generate a dad joke"
      }).pipe(Effect.provide(model))
  
      return { getDadJoke } as const
    })
  }) {}
  
  // The requirements are lifted into the service constructor
  //
  //          ┌─── Layer<DadJokes, never, OpenAiClient>
  //          ▼
  DadJokes.Default

Patch Changes

• Updated dependencies [ee0bd5d, 5189800, 58bfeaa, 194d748, 918c9ea, 9198e6f, 2a370bf, 58ccb91, fd47834]:
  • effect@3.16.0
  • @effect/experimental@0.48.0
  • @effect/platform@0.84.0