Cachet - AI Debugging CLI

Cachet sits between your app — where errors happen — and an LLM. It builds high-quality diagnostic prompts from runtime failures and learns from past fixes by storing resolved cases globally, so the same failure is never diagnosed from scratch twice.

Workflow

• Capture: cachet capture redacts secrets, fingerprints the route, and stores the failure locally — with zero network calls.
• Ask: cachet ask <id> assembles the diagnostic context and hands it to Anthropic or OpenAI.
• Proxy & watch: sit in front of a service (proxy) or tail traffic over ngrok (watch) to capture failures automatically.
• Memory: a 3-tier recall cascade (exact → fingerprint → embeddings) surfaces prior resolved cases.

Built for agents

Ships an MCP server and a skills.sh-compatible skill, so Claude Code, Cursor, and Windsurf can drive it in-editor. Installable globally via npm install -g cachet-cli.