Source control for coding agents: links each commit to the agent session that produced it and shares memory across Claude Code, Codex and other agents.
furkankly/zoetrope
A terminal and browser viewer that draws Claude Code and Codex sessions as a live flow graph of agents and tool calls.
- Stars
- 1,051
- Forks
- 73
- Open issues
- 13
- Last commit
- 4 days ago
What is it?
zoetrope reads the transcript files that Claude Code and Codex write for each session and draws the session as a graph: the main agent, the subagents it spawns, and the tools each one runs. The graph updates as the session progresses. A finished run can be replayed at a chosen speed, paced by the session's own timestamps, and a running session can be followed live.
The viewer is read-only and works from local files only, so transcripts do not leave the machine. It runs as a native terminal app (`zoe`) built on ratatui, and the same core is compiled to WebAssembly so the app can also run in a browser by dropping a transcript onto the page. A Herdr plugin can open the session belonging to a focused agent pane directly.
The timeline is indexed by event rather than wall-clock time, and seeking backwards restores the graph exactly as it stood at that point. Clicking an agent shows its spawning prompt, model, and tool calls with timings.
Key features
- 01Draws the main agent, its subagents, and workflow groups as a live graph.
- 02Shows status, current tool, tool count, and output tokens on each agent card.
- 03Replays finished sessions or follows running ones, with adjustable playback speed.
- 04Scrubbing backwards restores the graph as it was at that moment.
- 05Click an agent to inspect its spawning prompt, model, and tool calls.
- 06Headless `zoe inspect` prints the session tree without a TUI.
- 07Runs in the terminal or in the browser via WebAssembly.
Use cases
- Watching a long-running Claude Code or Codex session to see which subagents are active.
- Replaying a finished agent run to understand how it divided up and executed its work.
- Debugging an agent's tool usage or spawning decisions after the fact.
- Viewing agent sessions inside a Herdr pane alongside the running agent.
Topics
FAQ
- What is zoetrope?
- A terminal and browser viewer that draws Claude Code and Codex sessions as a live flow graph of agents and tool calls. zoetrope reads the transcript files that Claude Code and Codex write for each session and draws the session as a graph: the main agent, the subagents it spawns, and the tools each one runs. The graph updates as the session progresses. A finished run can be replayed at a chosen speed, paced by the session's own timestamps, and a running session can be followed live.
- Is zoetrope free to use? What license does it have?
- zoetrope is open source under the MIT license. Check the license terms before using it in commercial products.
- Is zoetrope still maintained?
- Actively maintained. The last commit to zoetrope was on Oct 14, 2026.
- What can I use zoetrope for?
- Watching a long-running Claude Code or Codex session to see which subagents are active. Replaying a finished agent run to understand how it divided up and executed its work. Debugging an agent's tool usage or spawning decisions after the fact. Viewing agent sessions inside a Herdr pane alongside the running agent.
- Where can I get zoetrope?
- The source code and installation instructions are on GitHub: https://github.com/furkankly/zoetrope
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A mobile, web and CLI front end for running Claude Code and Codex sessions from your phone, with end-to-end encrypted sync between devices.
Local proxy that lets OpenAI Codex and Claude Code run on any LLM provider, including Claude, Gemini, Grok, DeepSeek, and Ollama.
A terminal tool that replays Git commits as animated typing sessions with syntax highlighting and file tree transitions.
Terminal network diagnostics tool in Rust that shows live connections, throughput, packets and latency, and flags network issues.
A browser for AI agents where agents run automation in isolated Spaces while you keep browsing your own tabs and logins.