Coming Soon

A TypeScript runtime built for agents.

Fino makes AI a runtime subsystem, not a pile of glue: models, tools, agents, realms, and sandboxing — with OpenTelemetry built in from the start.

Join the DiscordExperimental

First Look

This is the whole app.

app.ts

import { agent, openai, streamText, tool } from 'fino:ai';
import { App } from 'fino:net/http/app';

const assistant = agent({
  model: openai({ model: 'gpt-4o' }),
  instructions: 'Answer as a concise support engineer.',
  tools: [lookupTicket],
});

const app = new App();

app.route('/chat').sse(async (events, ctx) => {
  const { message } = await ctx.request.json();
  for await (const text of streamText(assistant.stream(message))) {
    await events.write({ data: JSON.stringify(text) });
  }
});

app.listen({ port: 3000 });

The agent loop, tool validation, router, and streaming are all part of the runtime. fino run --watch app.ts and it's live.

Why Fino

Batteries included. Boundaries enforced.

AI is a subsystem

fino:ai ships provider-neutral models, agents, validated tools, durable sessions, MCP adapters, evals, and memory — built to work with routes, realms, and telemetry.

Imports are permissions

Capabilities come from explicit fino:* imports, not an ambient global bag. Child realms start from deny-all and can't grant themselves what a parent denied.

Realms and sandboxes

Isolated execution contexts run embedded, in threads, in processes, or remotely. Sandbox policies fail closed when the platform can't enforce them.

Coming Soon

Follow along.

Fino is experimental and moving fast. Hop into the Discord to follow the build, ask questions, and tell us what you're building.

Join the Discord