zoetrope — watch your Claude Code and Codex agents work
zoetropekeysdocsgithub ↗open app
╭ ──────────────────────── ┤ zoetrope ├ ───────────────────────── ╮ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ │ ╰────────────────────────────────────────────────────────────╯ ╭ main ─────────────────╮ │ │ ● opus-4 ↑ │ │ ❯ Edit ×14 ╰───────────────────────╯ ╭ tester ──────────╮ │ │ ✓ haiku │ │ ❯ Bash ×3 ╰──────────────────╯ ╭─ explorer ───────────╮ │ │ ● sonnet │ │ Grep ╰──────────────────────╯ 6.1k 6.4k 6.7k ◌ ✓ ×7 ×8 ▆ ▃ █ ▂▅ ▇ ▄ ▁ ▆▃ █ ▂ ▅▇▄ ▆ ▃ ▇ ▂ ▅ █▄ ▃ ▆ ▂▇ ▄ ▅ ▃ ▆ ▃ █ ▂▅ ▇ ▄ ▁ ▆▃ █ ▂ ▅▇▄ ▆ ▃ ▇ ▂ ▅ █▄ ▃ ▆ ▂▇ ▄ ▅ ▃ ▆ ▃ █ ▂▅ ▇ ▄ ▁ ▆▃ █ ▂ ▅▇▄ ▆ ▃ ▇ ▂ ▅ █▄ ▃ ▆ ▂▇ ▄ ▅ ▃ ├─────────────────────────────────────────────────────────┤ ❋ »»» ✗ ❋

watch your
agents work.

Claude Code and Codex write a transcript for every session. zoetrope reads it and lays the session out as a graph: which agents ran, and what each one is doing. Watch a live session, or replay a finished one.

reads the transcripts Claude Code and Codex already write · runs fully local.

●

the whole tree

Every agent and the subagents it spawns, on one graph. See who kicked off what, and what each is running now.

⚒

live tool activity

See each tool call as an agent makes it, and whether it passed or failed.

◆

time-travel any session

Scrub back, pause, jump between prompts, or snap back to live. The timeline is spaced by how busy the session was, not by the clock.

↳

full provenance

Click any agent to see why it ran: the prompt that spawned it, and the reasoning behind it.

the whole session · overview

One agent fans out to four subagents and a review workflow. The timeline fills as it goes — a ✗ marks the test run that failed before it was fixed.

following the work · live camera

Hand the camera to the action (f) and it glides to whichever agent just did something, at a readable zoom. This is what watching a live session looks like.

using it · pan, zoom, inspect, scrub
a Codex session · same app

The same graph from a Codex session: the CLI on the same brief, four agents spawned in parallel, and both layers of every exec call, the program the model wrote and the commands that actually ran.

Click an agent for its provenance — the prompt that spawned it, the reasoning, every tool it ran. Then drag the scrubber to travel back through the session.

The same engine compiles to WebAssembly —replay this exact session in your browser, no install.

~/code/my-app $ zoe█

see what your agents are doing.