kwant
Your own team of AI bots in one self-hosted binary: spaces, every account you have, handoffs, approvals and routines. Open-source, self-hosted alternative to Gr
What is kwant?
Your own team of AI bots in one self-hosted binary: spaces, every account you have, handoffs, approvals and routines. Open-source, self-hosted alternative to Grok Bot, Claude Duet and OpenAI dots. Rust + React.
Overview of kwant
What it is
kwant is a self-hosted binary that provides a team of AI bots sharing chats, handing work to each other via @mentions, and completing tasks autonomously or on a schedule. It is an open-source, self-hosted alternative to Grok Bot, Claude Duet, and OpenAI dots. The software is written in Rust (using Tokio, Axum, SQLite) with a React web app embedded in the binary. It can run as an app on macOS (Apple silicon and Intel) or Linux (x86_64 beta, .deb or AppImage), or on a server via binary or Docker. Users can connect existing AI engines such as Claude, Codex, or Grok CLI, or use their own API keys.
Key features
- Team of bots with distinct roles, voices, and models; bots can @mention each other to hand off work.
- Mission board that turns work into cards; a card becomes Verified only when kwant runs the check itself.
- Dev mode for code: shows what landed in each team's repo, reviews pull requests by a bot that didn't write the code, and requires approval for pushes to main, deploys, and other production steps.
- Spaces: separate environments (e.g., work, home) each with its own bots, teams, channels, routines, accounts, and approval rules.
- Multiple accounts per AI engine (Claude, Codex, Grok CLI, Anthropic, OpenAI, xAI keys, or custom servers) with priority order and automatic fallback.
- Routines on a schedule, follow-ups, and an Inbox for pending, finished, or failed tasks; Automation Discovery suggests routines for repeated tasks.
- Safety controls: default mode asks only on risky steps (e.g., git push, network calls); payments always ask; per-space or per-bot approval modes; Flight School grants freedom gradually.
Use cases
- Managing separate work and personal projects with dedicated bots and accounts.
- Code development with automated pull request reviews, risk assessment, and production safeguards.
- Automating routine tasks and follow-ups while the user is away.
- Handling bills and forms: kwant reads chats and IMAP inbox, extracts sender, ask, and deadline, and drafts a reply (never sends).
- Comparing deployment options via Parallel Universe: up to three ways with local container tests and dated price estimates.
Getting started
See the repository README for installation instructions.
License and pricing
The project is open-source under the AGPL-3.0 license.
Based on the project's public README and repository metadata.