Terminal user interface for AI agents built on agent-utilities.

A Textual-based terminal interface for interacting with AI agents. Connects to an agent-utilities backend via dual protocol support: AG-UI (SSE streaming) and ACP (JSON-RPC + SSE).

• Dual protocol support -- AG-UI (SSE streaming, default) and ACP (JSON-RPC + SSE, opt-in)
• Dynamic workflow sidebar -- discovers graph nodes from sideband events at runtime; nodes are never hardcoded
• Phase labels -- Planning, Discovery, Execution, Validation
• Completed node markers -- checkmarks on finished specialists
• Rich tool execution display -- extensible formatter system for rendering tool calls and results
• Human-in-the-loop tool approval -- modal for confirming sensitive tool calls before execution
• Multi-modal image attachment -- attach images to messages for visual reasoning
• Session management and chat history -- browse and resume previous conversations
• MCP server browser -- inspect connected MCP servers and their tools
• Slash commands -- /help, /clear, /mcp, /history, /image
• Unified specialist visibility -- MCP agents and A2A peers appear identically in the workflow sidebar; both emit the same sideband events via the backend's discover_all_specialists() unified roster
• Tool-count telemetry -- tools-bound sideband events include toolset_count, dev_tools, and mcp_tools breakdowns for per-specialist visibility

Start the agent-utilities backend server, then launch the TUI:

agent-tui

Or run with uv if installed locally:

uv run agent-tui

• uv
• prek

Create uv virtual environment and install dependencies:

uv sync --frozen --all-groups

Set up git hooks:

prek install

To update dependencies (updates the lock file):

uv sync --all-groups

Run formatting, linting, and type checking:

uv run ruff format && uv run ruff check --fix && uv run ty check