local-first · open source · mit
ThreadWatch is a reading room for AI coding sessions — Claude Code, Codex, Cursor, Copilot, and 25 more. Live-tail running sessions, inspect every tool call, and track token spend. Entirely on your machine.
also for Windows and Linux · install without the macOS dialog · checksums
what it does
Agents optimize for doing. ThreadWatch optimizes for understanding what was done.
Watch an active session update in real time — new messages append in place with a Live badge and a new-messages pill, no refreshing, no re-opening. When Claude Code is working, ThreadWatch is already reading.
The composer sends a prompt back into Claude Code (claude --resume) and streams the response into the transcript. Desktop-only by design — a LAN-exposed server can never become remote code execution.
Thinking, tool calls and results, diffs, images, citations, sub-agent trees, MCP calls — each content type has a purpose-built renderer, and virtualization keeps 10,000-message sessions scrolling at full speed.
Global search across every provider at once, per-session filters by role and content type, a message navigator for long transcripts, and export to HTML, JSON, or Markdown — including a headless --export CLI.
analytics
Model-aware cost estimates, cache-tier breakdowns, activity heatmaps, and tool-usage statistics — per project and global. Billing clarity your invoice never gives you.
one viewer for all of them
# build once, with the server feature cargo build --release --features webui-server # serve your history to any browser ./threadwatch --serve Auth token enabled: e60ed7c7... WebUI server running at http://0.0.0.0:3727
server mode
Run headless on a VPS or homelab and read from any browser. Token auth on by default, Docker image and systemd unit included, real-time updates over SSE.