ai

js/ai.ts

fino:ai — first-stop exports for Fino AI applications.

This module gathers the stable, application-facing AI APIs into one import surface. Use it when building agents, tools, sessions, memory-backed conversations, local or remote model calls, and evals. Specialized modules such as fino:ai/model, fino:ai/mcp, and fino:ai/context remain available when an application needs narrower imports or advanced types.

Design

The facade does not add behavior. It re-exports the same symbols from the underlying modules so users can start with one import and later split imports by subsystem without changing runtime semantics. Each symbol is documented in its home module; this page maps which subsystems the facade covers.

What's included

• fino:ai/agent — Agent, agent(), and streamText(): the high-level model loop that drives tools and strategy-owned history.
• fino:ai/model — provider-neutral Model messages and streams, plus the anthropic(), openai(), and local() adapters, provider factories, the ModelRegistry, and the model error types.
• fino:task — task() and Task: shared executable operations usable from CLIs, agents, and MCP, with toTaskToolDefinition() for tool calling.
• fino:ai/tool — tool() and Tool: validated tool definitions, plus SuspendSignal for human-in-the-loop pauses.
• fino:ai/runtime — GuardrailError, the maxSteps() stop condition, and the runContext async context populated while an agent step runs.
• fino:ai/cache — cachedModel(): exact and semantic response caching wrapped around any chat model.
• fino:ai/session — session() and Session: durable agent runs with session-owned history in an in-memory or SQLite store.
• fino:ai/memory — memory(), retriever(), and SqliteMemory: thread memory, working memory, and vector recall that outlive one context window.
• fino:ai/eval — evaluate() with scorers (exactMatch, contains, semanticSimilarity, llmJudge, schemaScorer) and reporters.
• fino:ai/skill — skill() and skillRegistry(): lazily loaded instructions, resources, and tools packaged behind a manifest.
• fino:ai/mcp — Model Context Protocol client and server adapters with stdio and HTTP transports, plus mountMcp() for serving over HTTP apps.

import { agent, openai, streamText, tool } from 'fino:ai';
import { v } from 'fino:validate';

const lookup = tool({
  name: 'lookup',
  description: 'Look up a record by id.',
  parameters: v.object({ id: v.string() }),
  execute: async ({ id }: { id: string }) => `record:${id}`,
});

const bot = agent({
  model: openai({ model: 'gpt-4o' }),
  tools: [lookup],
});

for await (const text of streamText(bot.stream('lookup A-1'))) {
  console.log(text);
}

No exported declarations found.