Comments
>log in to commentI built Agent Think Map as a visual debugger to help understand and troubleshoot AI agents. It lets you see runs, prompts, and hook flows in one place. Let me know what works for you and what could be improved 😊
Agent Think Map is a visual debugger that lets you explore and compare execution traces of AI coding agents.
- for
- Developers building or debugging Claude Code or Codex AI agents
- pricing
- open source
- license
- MIT
Key features
- Run History Search — Filter recorded runs by provider, status, outcome, date and label them.
- History Import — Load Claude Code transcripts without installing hooks, with dry-run preview.
- Trace Explorer — Navigate the execution graph and inspect input, output, errors and timings of each step.
- Run Comparison — Show exact or inferred differences between two runs with evidence view.
- Problem Detection — Identify repeated error patterns across runs and suggest troubleshooting checks.
- Agent Handoff View — Visualize parent-child agent relationships within a run.
- Local Workspace — Store runs in a local SQLite DB, preserving history across restarts.
- Embeddable Canvas — Render the map in any web app via a simple embed tag.
Use cases
- Debug a permissions error that caused a tool call to fail in a Claude Code session
- Compare two versions of a Codex agent to see what changed in their reasoning steps
- Identify recurring tool-call failures across many runs to prioritize fixes
- Inspect sub-agent delegation paths when an agent spawns another agent
- Sanitize and review captured prompts and tool data before sharing
Agent Think Map vs alternatives
 Agent Think Map |  LangChain |  LlamaIndex |  Mastra |  LangGraph | |
|---|---|---|---|---|---|
| Best for | Visual debugging of AI agent runs | LLM app framework | Data-aware LLM apps | TypeScript agent framework | Stateful agent orchestration |
| Pricing | Open source | Free | Free | Free | Free |
| DevHunt upvotes | 1 | 0 | 0 | 0 | 0 |
| Launched | Oct 2026 | — | — | — | — |
- Agent Think Map vs LangChain: LangChain is a code-first framework for building LLM apps, not a visual debugger.
- Agent Think Map vs LlamaIndex: LlamaIndex focuses on data-centric retrieval, not step-by-step trace visualization.
- Agent Think Map vs Mastra: Mastra is a TypeScript framework for building agents, lacking built-in debugging UI.
- Agent Think Map vs LangGraph: LangGraph provides stateful agent orchestration but no visual trace explorer.
Agent Think Map FAQ
Do I need an API key or account to use Agent Think Map?+
No, the hosted canvas demo and local Studio run without any account or API key.
Which Node version is required?+
Node.js version 22.13 or newer is required for the Studio.
Can I use Agent Think Map with both Claude and Codex agents?+
Yes, it provides separate install commands for Claude Code and Codex, each adding appropriate hooks.
Where is the trace data stored?+
Runs are stored in a local SQLite file at ~/.agent-think-map/runs.db.
How do I import existing Claude Code history?+
Run npx agent-think-map@latest claude --import-history with optional --dry-run to preview the import.
Summarized by DevHunt from github.com · Oct 15, 2026. Details may change; check the official site.
