Back to Open Source
Claude
MCP
Codex
React
MIT
Agent Think Map
Watch your AI agent think — inside the chat UI you already ship.
An embeddable canvas that draws an agent’s route as it runs: prompt → chain-of-thought → skill → tool → MCP → answer. Click any node to see why it fired, what went in, what came out, and how long it took. It’s a viewer, not a runtime — you keep Claude, Codex, LangGraph, or your own loop and emit one JSON object per step.
What you get
- Live canvas: prompt → thinking → skill → tool / MCP → subagent → answer, drawn as the run happens
- Click any node to inspect why it fired, its inputs, outputs, duration, and token or cost usage
- Scrubbable tool-call timeline for replaying a turn step by step
- One adapter ingests Claude Agent SDK, OpenAI Agents SDK, and Codex — or emit the JSON yourself
- Four lines of CDN HTML or a React component to embed; secrets in tool args are redacted
Who it’s for
Teams shipping AI agents and analytics copilots who need to debug the path a run took — which skill loaded, which tool ran, which MCP server answered — instead of reading a 4,000-line log after the fact.
Star it, fork it, use it
Open source under MIT — contributions welcome.