effect-ai-chat

Use Chat.fromPrompt or Chat.empty from effect/unstable/ai to create a session, then call session.generateText with each user message. The Chat module automatically appends prompts and responses to history, so multi-turn context is preserved without manual management.

Define tools with Tool.make and group them via Toolkit.make, then pass the toolkit option to generateText. Loop with an empty prompt until the model returns no tool calls; tool results are appended to history automatically.

Yes. Call session.exportJson to serialize the conversation, store the string anywhere, and restore it later with Chat.fromJson. For automatic saving after every generation, use Chat.layerPersisted with a BackingPersistence implementation.

Yes, session.streamText returns a Stream of response parts such as text-delta events. History updates when the stream finalizes, so consume the stream to completion if the full assistant reply should be recorded.

Every generateText, streamText, and generateObject call requires LanguageModel.LanguageModel in its context. Provide it per call with Effect.provide(modelLayer) or at the layer level when wiring your service.

Use Chat for any multi-turn conversation because it manages history accumulation and serializes generations with an internal semaphore. Use LanguageModel.generateText directly only for isolated single-shot generations with no shared context.