Open source, MIT

Your own AI assistant. On your own server.

A messenger where each contact is a bot that can browse the web, run commands, work on a schedule and remember you. It asks before it does anything you haven't allowed.

Why I built it

I built Tiny Agents to solve my own problems, and I think everyone should have a personal assistant they fully control.

Your data stays in SQLite files on your machine, you pick the model provider, and you decide what each bot may do without asking.

Ashik Nesin, @AshikNesin

What it looks like

A bot's Tools tab: each tool set to Allow, Ask or Off
Tools and permissions. Set each tool to Allow, Ask or Off, bot by bot.
Browser: a bot's Chrome, live
Browser. Watch a bot's Chrome live, or take over.
Computer: a bot's desktop, live
Computer. A full desktop the bots share with you.
Terminal: shells on the sandbox VM
Terminal. Shells on the sandbox that keep running.
Routines run a bot on a schedule
Routines. Run a bot on a schedule.

How it works

Set it up once. After that, you just talk to it.

1

Make it yours

Give it a name, a personality and a model. Decide what it may do on its own and what it asks about first.

2

Tell it what you need

In plain words. It already knows your preferences, so you don't repeat yourself, and it does the legwork while you get on with your day.

3

Say yes when it counts

Looking things up happens on its own. Paying, posting or changing anything waits for one tap from you.

What a bot can do

Ask before acting

Set each tool to Allow, Ask or Off. Approve from the web app, a push notification or the Android app.

Use a real browser and computer

A bot drives its own Chrome and, on a sandbox VM, a full desktop. You can watch live and take over.

Run on a schedule

Routines run a bot's prompt once, daily, weekly or on an interval, and report back.

Remember you

Per-bot memory, and a vault that fills logins and cards without the model seeing them.

Connect your tools

Add any remote MCP server.

Hand off code

The Code page drives coding agents (Claude Code, Shelley) and their pull requests.

Try it in two minutes.

No account with us and no hosted tier. You need Node.js 24 and pnpm 10.

git clone https://github.com/AshikNesin/tiny-agents.git
cd tiny-agents
pnpm install
TINY_AGENTS_FAUX=1 pnpm dev

The test model echoes you, so no API key is needed. Real models, servers and the sandbox are in the README.

  • RuntimeNode 24, one process
  • StorageSQLite
  • ModelsAnthropic, OpenAI, Gemini and more
  • Provider keysStay on your server or gateway
  • SandboxYour own VM or a Docker container
  • Sign-inPassword or passkey
  • Android appPlay Store soon, APK now
  • LicenseMIT

Questions

Is it one assistant or several?
Start with one. If you want, add specialists like a travel planner or a code helper, each with its own personality, model and tools. They can hand work to each other.
Can it spend money without asking?
No. A saved card asks for your OK every time it's used, and the model never sees the number.