Getting Started

muxd is an AI coding agent that lives in your terminal. It connects to LLM APIs (Anthropic Claude, OpenAI, Mistral, Grok/xAI, Google AI, Z.AI, Fireworks AI, DeepInfra, Ollama, and any OpenAI-compatible API) to help you read, write, and edit code through a streaming TUI interface.

Sessions are persisted in a local SQLite database so you can resume, branch, and search conversations across terminal sessions.

Architecture

muxd has three binaries. Run them all on one machine for local dev, or distribute them across servers for a team setup.

Binary	Role
`muxd`	Terminal TUI client
`muxd-daemon`	Agent server (sessions, tools, model calls)
`muxd-hub`	Hub coordinator (nodes, workers, web dashboard)

All three share one config file at ~/.config/muxd/config.json.

Install

Linux / macOS:

curl -fsSL https://raw.githubusercontent.com/batalabs/muxd/main/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/batalabs/muxd/main/install.ps1 | iex

From source (Go 1.25+):

git clone https://github.com/batalabs/muxd.git
cd muxd
go build -o /usr/local/bin/muxd .
go build -o /usr/local/bin/muxd-hub ./cmd/muxd-hub
go build -o /usr/local/bin/muxd-daemon ./cmd/muxd-daemon

Prerequisites: git (for undo/redo) and an API key for at least one provider.

Quick start

One machine: daemon + TUI

Everything local. Two terminals, no hub.

# Terminal 1: start the daemon
muxd-daemon

# Terminal 2: start the TUI (auto-discovers the daemon)
muxd

Set your API key inside the TUI:

/config set anthropic.api_key <your-key>

With a hub: multi-machine

Hub setup guide →

# Server: start the hub (prints token + QR code)
MUXD_HUB_BIND=0.0.0.0 muxd-hub

# Server: start the daemon (registers with hub)
MUXD_HUB_URL=http://localhost:4097 \
MUXD_HUB_NODE_TOKEN=<token> \
MUXD_BIND=0.0.0.0 \
muxd-daemon

# Your laptop: connect to the hub
muxd --remote <server-ip>:4097 --token <token>

Open http://<server-ip>:4097 in your browser for the web dashboard.

What's in the docs

Client setup — local daemon, CLI flags, TUI features, service install
Hub setup — multi-machine coordination, web UI, workers, Docker
Commands — full slash command reference
Configuration — API keys, model aliases, env vars, pricing
Tools — all 41 built-in agent tools
Features — deep-dives on activity, permissions/transfers, memory, hub dispatch, and more
MCP Servers — extend the agent with external tool servers
Architecture — codebase internals for contributors
Undo / Redo — how git-based checkpoints work
Project Memory — persist project facts across sessions
Contributing — build, test, and code style guide