In development

The full-stack framework for AI agents.

The agent on your server, the chat in your UI, and your data in between, acting with your users' permissions. One package, on any model, inside the app you already have.

Get started

your-app/

You are the support agent for Acme.

Look orders up before answering; never guess a status.
Offer a refund only when an order is more than
10 days late. Keep replies short.

The agent, its tools, the route and the UI that talks to it, in one project.

From the model to the message box.

Server to screen, one package

defineAgent on the server, useAgent in React, and a single route between them. Streaming, tool calls and thread history work end to end with no glue code.

Agents that know who's asking

Every tool runs with the signed-in user, the thread and the request in hand. Threads are scoped per user, so one person's conversation never leaks to another.

Your data, on your users' terms

Connectors run queries with the user's own credentials, so your database permissions still apply. Tables are allowlisted and writes are opt-in.

Any model, no middleman

Built on the AI SDK. Use any provider it supports, or your own. Model calls go straight to the provider you chose.

Runs where your app runs

Agents are web-standard Request and Response handlers: Next.js, Hono, Bun, Deno and edge runtimes, in the app you already have.

Less to wire, less to learn

A handful of functions and a folder per agent. Loom generates the wiring as you work, and stays out of the way when you need the AI SDK directly.

Small on purpose.

Loom gets an agent from your server to your users and stops there. No workflow engine, no hosted platform to adopt. Everything else you build on top with the AI SDK and the tools you already use, and anything you don't import is never bundled.

In Loom

  • Agent runtime and streaming
  • One route for every agent
  • React chat hook
  • Per-user threads and auth
  • Data connectors
  • Agent folders and codegen

On top of Loom

  • Retrieval and embeddings
  • Evals
  • Workflow engines
  • Voice

Bring your stack.

Models

  • Anthropic
  • OpenAI
  • Google
  • Mistral
  • OpenRouter
  • Ollama
  • Your own

Data

  • Supabase
  • Your own connectors

Frameworks

  • Next.js
  • Hono
  • Bun
  • Deno
  • Edge runtimes

UI

  • React

Ship an agent your users can actually talk to.

Read the quickstart