Watch & live graph

Evolve command (recommended)

codegenome evolve runs an initial build, starts a local HTTP server for the HTML graph, watches .py files, and optionally broadcasts deltas over WebSocket:

codegenome evolve --live --lan .

• HTTP UI (Local): http://localhost:8000/graph.html?live=1
• HTTP UI (LAN): Printed in terminal (e.g., http://192.168.1.x:8000/graph.html?live=1)
• WebSocket (with --live): ws://127.0.0.1:8765 (or LAN IP with --lan)

Live Graph AI Chat

When viewing the live graph in your browser, you can open the built-in AI Chat panel to query your architecture directly. Supported providers include Ollama (Local/Cloud), OpenAI, Gemini, and Groq. You can adjust the graph context size (minimal to max) to control how much local neighborhood data is sent with each prompt.

Legacy watch mode

The flag-based CLI supports debounced rebuilds and metric-based live graph polling:

# Rebuild after 10s of inactivity
python -m codegenome --workspace . --build --watch --watch-debounce 10

# Rebuild when file/line counts increase
python -m codegenome --workspace . --build --live-graph --live-graph-interval 60

MCP + watch together

For AI agents that need a continuously updated graph:

python -m codegenome --workspace . --build --mcp --watch

HTTP MCP listens on 127.0.0.1:7331 by default. See MCP server.

When to use which mode