TAU Documentation
TAU is a Rust-native AI coding agent built for maximum performance. Terminal-first, 75+ LLM providers, local embeddings, and native vision support.
Quick Start
# Install TAU (macOS)
curl -fsSL https://taulepton.com/install | bash
# Option 1: Use existing subscription (no extra cost)
tau auth login claude-code # Claude Pro/Max
tau auth login github-copilot # GitHub Copilot
tau auth login gemini-cli # Google AI
# Option 2: Use API key
export ANTHROPIC_API_KEY=sk-ant-xxx
# Run TAU
tauCore Features
Terminal UI
Ratatui-powered interface with live streaming, syntax highlighting, and keyboard shortcuts.
Learn more75+ LLM Providers
Anthropic, OpenAI, Google, DeepSeek, Groq, Ollama, and 70+ more via models.dev.
Learn more37+ Built-in Tools
File operations, code search, git, bash, LSP, DAP debugging, browser automation.
Learn moreNative Vision
Local MLX-powered VLM on Apple Silicon. Browser screenshots and image analysis.
Learn moreRLM (10M+ tokens)
Recursive Language Models for analyzing massive codebases. Fixed cost ~$0.47.
Learn morePermissions
Pattern-based rules with risk levels. Control what the agent can and cannot do.
Learn moreArchitecture
TAU is built as a Rust workspace with 18 specialized crates:
tau/ ├── tau-cli # CLI binary and entry point ├── tau-tui # Terminal UI (Ratatui) ├── tau-server # HTTP API + SSE server ├── tau-core # Runtime, LSP, Shell, Indexing ├── tau-llm # LLM providers (Anthropic, OpenAI, etc.) ├── tau-tools # 37+ built-in tools ├── tau-mcp # MCP client/server integration ├── tau-storage # SQLite + FTS5 persistence ├── tau-permissions# Risk-based permission engine ├── tau-skills # Skills system ├── tau-tasks # Task management + Knowledge Graph ├── tau-agents # Subagents (RLM, Plan, etc.) ├── tau-auth # OAuth + keyring storage ├── tau-acp # Agent Client Protocol └── ... # Additional crates
Performance
<2s
Cold start
<100ms
Tool execution
<100MB
Memory (idle)
24h+
No memory leaks