Your agents, one inbox.
Taskrun is one inbox for all your AI coding agents. When one needs an answer from you, it shows up here. Click your answer, and it gets back to work.
Your coding agents are working in the background. The dots on the right show each one: blue means working.
It's a real demo: click anything, or use your keyboard. 1 2 3 answers, Esc undoes.
The problem, in ten seconds.
You run AI coding agents
Tools like Claude Code and Codex write code for you, each in its own window.
They stop and ask you things
Mid-task, an agent needs a decision. It waits, buried in a window you forgot about.
Taskrun puts it all in one inbox
Every question lands in one place. You answer in a click, and every agent keeps moving.
Set up in two minutes. Then just answer.
Your agents already stop and wait for you. This gives the waiting a single place to end.
- Looking for agents0 connected
- Claude Code
- Codex
- OpenCode
- Amp
1Point your agents at it
One line in an agent's instructions, or a hook. It finds the agents you already run.
- 1 of 3 linked. Tap to toggle.2
Link your machines
Your Mac, your phone and any server you run agents on end up in the same inbox.
- claude-7Which layout?3
Answer, or just talk
Press a number in the inbox, or say what you need and it goes to the right agent.
taskrun ask "Run migration 0042?" --option "Yes" --option "Skip"It keeps track, so you don't.
Small things, done carefully. Try them.
Questions.
Which agents does it work with?
Anything that can run a shell command, so Claude Code, Codex, OpenCode, Amp, Pi, Grok or your own scripts all work the same way.
When can I try it?
Taskrun is still being built. Leave your email below and we'll write once, when early access opens.
Where does my data go?
Nowhere. Taskrun is built to run on your own machine, so your agents' questions and your answers stay with you.
Does it act on my agents by itself?
No. It only passes along what you answer or ask for. Every answer waits two seconds first, and Esc takes it back.
How does routing work?
It matches your request against each agent's name, project and what it last said it was doing, and picks the closest. If nothing matches, it tells you instead of guessing.
Answer once.
Then let them get on with it.
Nothing to sign up for yet. Just your email, once.