Corvade
See what your agents see
Overview
A local AI agent control plane that sits between your agents and LLM APIs as a transparent proxy, capturing every call and visualizing execution as a real-time graph. Think Wireshark for AI agents.
Architecture
Acts as an HTTP reverse proxy that intercepts every LLM API call from your agents. Each call is logged, timestamped, and pushed via WebSocket to a live graph UI. The graph shows agent decision trees — branching tool calls, retries, parallel executions — rendered as an interactive node graph.
The Story
Inspired by watching Claude Code debug a complex refactor across 20 files and having no idea what it was thinking. The transparent proxy approach was the simplest path — no agent modifications, just redirect your API endpoint. The graph visualization came from a late-night experiment with d3.js force-directed layouts that accidentally looked incredible.
Technology
Links
Features
- Transparent proxy between agents and any LLM API
- Real-time execution graph visualization
- Full prompt and response capture
- Token usage tracking per call
- Local-first — all data stays on your machine
Highlights
- Companion to Vexade (pre-production agent dev platform)
- Works with any OpenAI-compatible API