Quiet, live, read-only timelines for local coding-agent sessions.
Each observer reads the transcripts its agent already writes and shows public messages, commands, file operations, MCP calls, subagents and tool results on the same twelve-row timeline. One shared local process per agent. No additional model calls. No API key.
| Folder | For | Installs as | Status |
|---|---|---|---|
codex/ |
Codex desktop and CLI | codex-trajectory |
v0.3.0 preview |
claude/ |
Claude Code (desktop and CLI) | claude-trajectory |
0.1.x preview |
Codex
Claude Code
Both screenshots use fictional demonstration data, not real conversations. Click a preview for that observer's setup and usage instructions.
The two implementations are independent: separate installers, hooks, runtime directories, URL schemes and recovery apps. Installing one never changes the other. Run each folder's commands from inside that folder, for example:
git clone https://github.com/Fly-Carrot/codex-trajectory.git trajectory
cd trajectory/claude
python3 install.py --dry-runShared fixes are applied to each folder separately and tested with that folder's own
suite. Tags are prefixed by component (codex-v…, claude-v…); earlier v0.x tags
belong to the Codex observer.
MIT licensed. See SECURITY.md.

