claudectl

claudectl manages multiple Claude Code sessions for developers who need to track parallel work and intervene when an agent gets stuck. Its terminal dashboard brings session activity, approvals and spending into one view.

• Session oversight: See status, context use, CPU and memory consumption, then approve requests, send input or stop a session from the dashboard.
• Spending and health controls: Enforce session or daily spending limits, detect stalls, repeated errors and context saturation, and inspect completed sessions for failures and costs.
• Local LLM supervision: An optional model approves or denies tool calls, routes summaries between sessions and starts parallel work, learning from manual decisions and corrections.
• Agent coordination: Dependency ordering and file ownership controls help agents avoid duplicate work, while peer-to-peer connections share learned preferences and delegate tasks across machines.
• Decision review: Inspect the reasoning, confidence and past examples behind supervisory decisions, correct mistakes and keep separate preferences for different projects.

The manager runs locally on macOS and Linux and is open source under MIT. Local monitoring needs no API key, authentication or network access. The supervisor supports Ollama, llama.cpp, vLLM and LM Studio through OpenAI-compatible endpoints; learned preferences stay on your machine. Configured webhooks can send data to external services.

A Claude Code plugin brings supervision into coding sessions. Headless operation keeps monitoring and coordination active without the dashboard. Terminal support varies: tmux and Kitty support launching, switching, input and approvals, while other terminals support fewer controls.