tau-coding-agent 0.5.5

A self-extensible agent CLI with terminal UI, multi-provider LLM support, session management with branching, and a plugin system for tools and commands

Tau (τ)

Inspired by Pi, Tau brings the same agent framework capabilities to Python developers. A self-extensible agent CLI with a terminal UI, multi-provider LLM support, session management with branching, and a plugin system for tools, commands, and customization.

Start a conversation with the agent in the terminal.

---

Quick Start

Installation

uv pip install tau-coding-agent

Or install from source:

git clone https://github.com/Jeomon/Tau.git
cd Tau
pip install -e .

Launch

tau

Authenticate with your LLM provider using /login or set environment variables (e.g., NVIDIA_API_KEY).

Common Commands

tau                                           # Interactive mode
tau --resume                                  # Resume most recent session
tau --model claude-sonnet-4-6                 # Use specific model
tau --print "Summarize this repo"             # One-shot mode
tau --provider anthropic --print "Explain this code"  # Print mode with provider
tau --mode rpc                                # RPC mode for IDE extensions

For detailed options, see CLI Reference

Features

• Multi-provider LLM support — Anthropic, OpenAI, Google Gemini, Mistral AI, Ollama, Groq, xAI, AWS Bedrock, OpenRouter, Z.ai, and more
• Terminal UI — Built-in chat interface with syntax highlighting, markdown rendering, and keyboard navigation
• Rich media support — Work with text files, images, audio files, and video files via file references or clipboard input
• Session management — Persistent sessions with branching, forking, and resuming capabilities
• Tool execution — Built-in tools (terminal, file I/O, search) with full user permissions and extensibility
• Plugin system — Add custom tools, slash commands, hooks, themes, and skills without modifying core code
• Themes — Dark and light base themes, customizable and extensible via YAML
• Context management — Automatic context compaction and branch summarization for long conversations
• Python API — Embed Tau in your own applications programmatically
• Multiple run modes — Interactive TUI, print mode (one-shot), JSON event stream, and JSON-RPC for IDE integration

Documentation

Start here: Tau Documentation

Key resources:

• Quickstart — Five-minute getting started guide
• Installation & Setup — Provider authentication and configuration
• Usage Guide — Interactive mode and slash commands
• Architecture — System design and data flow
• Extensions — Building tools, commands, and plugins
• Python API — Programmatic usage

Core Architecture

Console (CLI) → TUI (Terminal UI) → Runtime (Agent Execution) → Engine (Tools) → Inference (LLM)

User input flows through the TUI to the runtime, which executes tools via the engine and calls the LLM provider for inference. Results are rendered back in the TUI and persisted to sessions.

Configuration

Settings are loaded in order of precedence:

1. Built-in defaults
2. ~/.tau/settings.json (global user settings)
3. .tau/settings.json (project-level settings)
4. Environment variables
5. Command-line flags

Sessions are saved to ~/.tau/sessions/ and can be resumed, forked, or cloned.

Project Context Files

Tau automatically discovers and includes project-specific instructions from AGENTS.md or CLAUDE.md in the system prompt. This allows you to:

• Define project rules and coding guidelines for the agent
• Standardize how the agent handles project-specific conventions
• Store project context separately from tool configuration

Example usage:

# Auto-discover AGENTS.md in the project (default behavior)
tau

# Disable project context file loading
tau --no-context-files

# Trust project files explicitly
tau --approve

# Don't trust project files for this run
tau --no-approve

See Project Context Files for detailed instructions.

Security & Permissions

Tau executes tools (terminal, file I/O, web operations) with the permissions of the user and process that launched it. There is no built-in permission system for restricting filesystem, process, network, or credential access.

If you need stronger boundaries, consider:

• Running Tau inside a container or sandbox
• Using OS-level security policies
• Configuring environment variables to limit tool access

Supply Chain Security

We treat dependency changes as reviewed code changes with these practices:

• Exact version pinning — All direct dependencies pinned to specific versions in pyproject.toml
• Lockfile integrity — uv.lock is the source of truth; changes require explicit review
• Dependency auditing — Use pip-audit or safety to scan for known vulnerabilities
• Safe installation — Install with --no-deps to prevent malicious lifecycle scripts

See SECURITY.md for detailed practices and vulnerability reporting.

Development

python -m pytest              # Run tests
pyright tau/                   # Type checking
python -m tau                  # Launch from source

See Development Setup for detailed instructions.

Contributing

See CONTRIBUTING.md for contribution guidelines and AGENTS.md for project-specific rules.

License

This project is licensed under the MIT License — see the LICENSE file for details.