onsails/right-agent

<p align="center"> <a href="LICENSE"><img src="https://img.shields.io/badge/license-Apache--2.0-E8632A.svg" alt="license"></a> <a href="https://github.com/onsails/right-agent/actions"><img src="https://github.com/onsails/right-agent/actions/workflows/build.yml/badge.svg" alt="build"></a> <a href="https://t.me/rightagent"><img src="https://img.shields.io/badge/Telegram-chat-E8632A?logo=telegram" alt="telegram"></a> </p>

<img src="assets/section-mark.svg" height="20" alt=""> right agent

right agent is an ai agent you run by messaging it. you can give it real credentials without handing them to the model – every agent runs in its own sandbox; every credential lives outside it. the secret bytes never enter the box, so the worst a compromised agent can do is misuse a tool while it runs. it can't read the credential, and the proxy won't resolve it onto the open internet. for anyone tired of "grant all permissions and hope," that is the change.

<img src="images/screenshot.png" alt="Right Agent in Telegram" width="720"/> <p align="center"> <a href="#-what-you-get">what you get</a> · <a href="#-how-it-works">how it works</a> · <a href="#-install">install</a> · <a href="#-security">security</a> · <a href="#-how-it-compares">how it compares</a> · <a href="#-roadmap">roadmap</a> · <a href="#-docs">docs</a> </p>

one Telegram bot per agent. each chat – a dm, a group, a topic inside a group – is its own Claude Code session over shared, chat-tagged memory. so a dm and a group topic are different working contexts, but they remember the same things about you. you talk to an agent in Telegram; that's the whole product.

it remembers what matters, and it can act on your behalf without you handing it the keys to your machine. the choices are already made – sandboxed by default, Telegram is your only console, many agents on one Claude subscription. the box is closed; you just use it.

[!WARNING] **today every agent runs on Claude Code (claude -p), so you