See running Claude Code agents on your Mac, in real time. Replay finished sessions. Understand what your agents actually did.
AgentHub is a small, local-first desktop app. No telemetry, no network — it just watches ~/.claude/ and shows you what's happening.
- Live agent dashboard — every running Claude Code session on your machine, with current todo list, tool usage, and tokens spent.
- Session replay — scrub through any finished session's full transcript: prompts, tool calls, file edits, sub-agent dispatches.
- Sub-agent tree view — when an agent dispatches sub-agents (Task tool), see the tree, drill into each branch.
- Cost & token roll-ups — per session, per project, per day.
(Features ship one at a time via the autonomous harness. See docs/ROADMAP.md for what's next, and ~/workspace/Private/docs/Projects/Agent-Viz/shipped/ for hero shots of what's already landed.)
Long agent runs are hard to review. The terminal scrollback is gone, the .jsonl transcripts are dense, and you can't easily tell which sub-agent did what. AgentHub turns that data into something you can actually look at.
- Tauri 2 (Rust) — native shell + filesystem access, no Electron weight
- React 19 + TypeScript + Vite — UI
- Bun — package manager + script runner
Pure local file watching against ~/.claude/sessions/, ~/.claude/projects/, ~/.claude/todos/. Nothing leaves your machine.
bun install
bun run tauri devbun run tauri buildThe .app / .dmg lands in src-tauri/target/release/bundle/.
AgentHub is itself developed by an autonomous Claude Code harness — a virtual team of subagents (PM, Designer, Dev, QA, Reviewer) coordinated by a Lead session. See CLAUDE.md for the operational protocol.
MIT