Thrum

I built Thrum so you can run several AI coding agents in parallel without becoming the message relay yourself. You do the thinking — research, plan, approve. Agents do the typing. Thrum routes messages between them, keeps sessions alive, and stays out of your way. It doesn't plan your work or make decisions for you. That's a deliberate choice.

What You Can Build With It

Here are the four main shapes of how people use Thrum today. Each one is a complete workflow, not a feature list. Pick the one that matches where you are right now — solo with one agent, a local team of agents, agents spread across repos or machines, or fully automated plan execution. They're not mutually exclusive. Most people start with the first and add complexity only when they need it.

>_

Solo Dev with One Agent

One agent, your machine, no hand-holding. Thrum keeps the session alive in tmux, tracks identity across context resets, and lets you check in from your phone via Telegram.

Walk through the setup →
@

Team on Your Machine

Two, three, or ten agents in parallel worktrees. A coordinator plans, implementers build, a tester verifies. You review and merge. Messaging runs locally; Telegram works here too.

Walk through the setup →

Agents Across Repos/Machines

Backend agents in one repo talk to frontend agents in another. Your home desktop and your work laptop participate in one team mesh. Same-subnet or Tailscale — your pick.

Walk through the setup →

Automated Plan Execution

You wrote a plan. Hand it to the orchestrator. It spins up implementer agents, runs them epic by epic, stops at review gates, and hands you a merge report. You still merge.

Walk through the setup →

Further Reading