You already have todo.md. You already have activity-log.md. Every coordination tool I tried wanted another copy- import them, sync them, store them somewhere else.
Dispatch reads your files as-is. Dispatch agents from a dashboard, watch them work in a terminal, validate the output. Your markdown stays the source of truth. The dashboard is a lens over it.
Experimental. YOLO mode on by default. Working prototype, not a finished product. Rough edges guaranteed.
Not a chat interface. Claude.ai and ChatGPT are for conversations. Dispatch is for work. The unit is a task with a defined outcome, not a message thread you scroll back through.
Not an agent. Dispatch doesn't write code or run commands itself. It coordinates the agents that do- Claude Code, Codex, OpenClaw. You still pick the agent and review the work.
Not OpenClaw. OpenClaw is an agent gateway- it runs models and manages their permissions. Dispatch is the coordination layer on top. They complement each other; Dispatch supports OpenClaw as an agent option.
Not a task manager. Linear, Jira, and GitHub Issues are databases with task-shaped views. They want to own your work history. Dispatch reads your existing todo.md and writes back to it. You can stop using Dispatch today and nothing is lost.
Not Devin or Copilot Workspace. Those are cloud platforms that manage agents in their infrastructure. Dispatch runs locally, coordinates agents in your terminal, and stores everything in files you control.
- Doesn't touch your existing setup
- Markdown is the truth. No new file formats, no database, no sync
- One agent per task, many tasks in parallel — each dispatched task gets its own agent session. Enable worktrees to isolate branches and run jobs concurrently without git conflicts
The CLI (cli.js, parsers.js) has zero third-party dependencies- only Node.js built-ins (fs, path, child_process). No npm supply chain exposure for the coordination layer.
The dashboard has 15 production dependencies: React, Express, xterm.js (terminal emulator), node-pty (pseudo-terminal), and supporting utilities. Nothing phones home. No accounts, no telemetry, no API keys stored by Dispatch.
The server binds to localhost only. It is not safe to expose to a network.
node-pty compiles a native binary during install- it's the one step that requires build tools. If yarn install fails, install python3 and Xcode Command Line Tools (macOS) or build-essential (Linux) first.
- Task dashboard — open tasks and activity across all your repos in one place
- Agent dispatch — start Claude Code sessions from the dashboard, watch terminal output live
- Auto-completion — validate a job and it marks the task done in
todo.md, logs it toactivity-log.md - Job history — every agent run logged to a markdown file in
.dispatch/jobs/ - CLI — JSON output for scripting and agent-to-agent queries
1. Write a task
Add a task to todo.md in any connected repo:
- [ ] Change the app logo to greenOr add it from the dashboard.
2. Dispatch from the dashboard
Click Start on a task to open the pre-filled dispatch form. Review, edit, send.
3. Agent works
A terminal opens in that repo with your prompt. The agent writes incremental progress to the job file as it goes. Watch live or come back later.
4. Review
When the agent finishes, the job moves to the review queue. Open it to read the progress log and see what changed.
5. Validate
Click Validate. Dispatch:
- Marks the task done in
todo.md - Logs an entry to
activity-log.md
The job file in .dispatch/jobs/ stays as a permanent record.
Nothing globally.
In the Dispatch directory:
| What | Path |
|---|---|
| Dashboard (React + Express) | dashboard/ |
| CLI + parsers | cli.js, parsers.js |
Claude skills (/dispatch, /add-repo, /done) |
.claude/skills/ |
In each repo you connect (via /add-repo):
| What | Path |
|---|---|
hub-stop.js hook |
.claude/hooks/ |
protect-env.js hook |
.claude/hooks/ |
| Hook registration | .claude/settings.json (merged with any existing config) |
todo.md, bugs.md, activity-log.md |
repo root (only created if missing) |
hub-stop.js is a no-op unless Dispatch dispatched that Claude session. Normal Claude usage in your repos is unaffected.
tmux keeps agent sessions alive across dashboard restarts. Without it, running Claude sessions die whenever the dashboard restarts.
# macOS
brew install tmux
# Ubuntu/Debian
sudo apt install tmuxDispatch uses tmux automatically when available. It works without tmux- sessions just won't survive restarts.
git clone https://github.com/pi0neerpat/dispatch
cd dispatch/dashboard && yarn install && cd ..claude .In the Claude Code conversation:
/add-repo
Claude asks for the repo name, path, and optional scripts, then sets everything up. Repeat for each repo you want to track.
cd dashboard && yarn devOpen http://localhost:5173.
cd dashboard && yarn test- Open tasks in your repos'
todo.mdfiles appear in the dashboard - Click a task to dispatch a Claude Code agent- it opens a terminal session in that repo
- The agent writes progress to
.dispatch/jobs/YYYY-MM-DD-slug.mdas it works - Validate the job and Dispatch marks it done in
todo.md, logs it toactivity-log.md
The dashboard reads your files. It doesn't own them.
Dispatch reads and writes standard markdown. If you already have these files, they're picked up as-is.
## Open
- [ ] Add input validation to the API
- [ ] Write tests for auth module
## Done
- [x] Set up CI pipeline## Open
- [ ] Login fails on Safari
## Fixed
- [x] 500 error on empty form submitUpdated automatically when you validate a job. Also writable via /done.
# My Repo — Activity Log
**Current stage:** MVP
## 2026-03-24
- **Added auth module** — JWT-based auth with refresh tokensnode cli.js status # Overview of all repos
node cli.js tasks # All open tasks across repos
node cli.js tasks --repo=app # Tasks for one repo
node cli.js swarm # Active and recent agent jobs (alias for jobs)
node cli.js repos # List configured repos
node cli.js config # Dump full resolved configAll output is JSON- designed for scripting and agent consumption.
Three skills are active when Claude Code is opened in the Dispatch directory:
| Skill | Invoke | What it does |
|---|---|---|
/dispatch |
/dispatch or "what should I work on?" |
CLI view across all your markdown files- tasks, activity, what to work on next |
/add-repo |
/add-repo |
Connect a new repo to Dispatch |
/done |
/done or "we're done" |
Mark a task done and log activity manually |
/done is for work done outside the dashboard- manual sessions, ad-hoc fixes, anything not dispatched through the UI.
- Better agent lifecycle management (tmux persistence is working; orphan cleanup is ongoing)
- Support for other agents: Codex, Cursor, etc.