AI coding agent.
Built in Rust. MIT licensed. Your code never leaves your machine.
curl -sSL https://raw.githubusercontent.com/rust-infra/tact/main/scripts/install.sh | bash ~15MB, no Electron, no Node.js. One binary, instant start.
Your API calls go direct. No cloud relay, no telemetry by default.
Truly open source. Fork, modify, redistribute without restrictions.
MCP servers · Custom skills · Hook system · 40+ built-in tools
Features
Agent loop
Streaming outputs, automatic context compaction, crash recovery. Run headless or interactive.
40+ tools
File I/O, Shell, LSP, Web search, Team messaging, Worktree isolation, Cron scheduling.
Permission modes
Choose your safety level: default (ask per action), plan (review before execution), auto (fully autonomous).
Sub-agents & worktrees
Spawn child agents in isolated git worktrees for parallel, sandboxed task execution.
Native MCP
Use any MCP server as a tool provider. Extend tact with the growing MCP ecosystem.
Quick Start
Install
Run the install script on Unix or Windows.
curl -sSL https://raw.githubusercontent.com/rust-infra/tact/main/scripts/install.sh | bash powershell -c "irm https://raw.githubusercontent.com/rust-infra/tact/main/scripts/install.ps1 | iex" Configure
Create a tact.toml in your project root.
[llm]
provider = "deepseek"
[llm.providers.deepseek]
api_key = "$DEEPSEEK_API_KEY"
base_url = "https://api.deepseek.com"
model = "deepseek-v4-flash"
models = [
"deepseek-v4-flash",
"deepseek-v4-pro",
]Run
Launch interactive or headless mode.
tact-ui --config=tact.toml tact-ui --config=tact.toml headless "Refactor this module"