Docs

Quickstart

Go from a fresh clone to a team of bots answering in your browser. On most machines it takes a few minutes, plus a one-time model download in the background.

What you need

Rust (stable), installed with rustup.
Node.js 20 or newer with npm, to build the web app. The finished binary doesn't need it.
At least one AI account for your bots. Any one of these is enough:
- the claude CLI, installed. You don't need to sign it in first, because kwant has a Sign in with Claude button;
- the codex or grok CLI, signed in on the same machine (codex login, grok login);
- an Anthropic, OpenAI or xAI API key, which you add in the app under Space settings → Accounts.
Optional: Docker for VM mode and the bots' shared browser, and uv or python3 for the Laya router.

1. Run from a fresh clone

git clone https://github.com/alexsergivan/kwant.git
cd kwant
make run   # builds the web app, then runs `cargo run -- serve`

Open http://127.0.0.1:8799. On your own computer there's nothing to sign in to. If no account works yet, kwant shows a Get ready banner with an Add an account button.

The first start takes a few minutes. kwant installs its Laya router and downloads its model (about 420 MB) in the background. Bots answer in the meantime, picked by keyword. To skip Laya, run cargo run -- serve --no-laya.

cargo run -- serve on its own works only once frontend/dist exists, because the binary embeds the web app. After the first make run (or make frontend) you can use either.

2. Add an account

Pick whichever you have:

Claude: click Sign in with Claude, open the link and paste the code shown back into kwant.
Codex or Grok: sign the CLI in on this machine, and kwant uses it as is.
API key: open Space settings → Accounts and add an Anthropic, OpenAI or xAI key. Keys are write-only: the app never shows them again.

You can add several accounts per engine, such as a work and a personal one. See Accounts & profiles.

3. Ask your first question

A fresh install's Default space starts with three bots in #general, on the @core-engineering team:

Bot	Title	Good at
Architect	Systems Architect	Architecture, software design, multi-file engineering
TechLead	Staff Engineer & Prototyper	Fast reasoning, prototyping, clear explanations
DevOps	Systems & Security Engineer	Shell, servers, containers, security

Type a question in #general. kwant picks the bot that fits best and shows why. Mention a bot by name (@Architect) to ask it directly, or ask in the team's chat (@core-engineering): when a question needs several views, the lead runs a short discussion and posts one summary.

A question in the general channel routed to Architect, with the routing reason and the model and effort used. — kwant shows which bot it picked and why. Screens on this site come from a sample space.

Next steps

Create bots and teams from ready-made personas.
Choose how often bots ask before running commands.
Run kwant on a server with systemd or Docker Compose.
Give bots tools like GitHub or Notion through MCP.

Running the tests

cargo test               # Rust tests (run `make frontend` first: they need frontend/dist)
cd frontend && npm test  # web app tests