Decide what’s worth building. Agents do the rest.

Relay is free to use, and it’s the tool we ship our own software with. You say what you want and approve the plan. AI agents write the code, review it and open the pull requests, and they ask you only when a decision is yours to make.

Relay’s What overview for a payments project: the outcome, who it’s for, the problem with its figures, and Pythia waiting for an answer before the member approves.

From an idea to shipped work

  1. 1

    Say what you want

    Describe the problem or the opportunity in your own words, with whatever you have to hand. Pythia, Relay’s planning agent, asks what she needs to know, one question at a time.

  2. 2

    Approve the plan

    Pythia writes up the problem, the outcome and how you’ll know it worked, and marks what she assumed so you can correct it. You approve it or ask for changes.

  3. 3

    Agents build it in milestones

    Pythia breaks the plan into milestones, biggest risk first. Athena, the building agent, builds each one and opens pull requests, and Artemis reviews them before they ship.

The payments project’s four milestones in the order Pythia proposed them, each saying what it delivers and what it waits for, with her explanation and the approve button beside them.

Free to use

You choose who runs the agents.

Bring your own agent

Free

Run the agents on your own machine, with your Claude subscription or API key.

Use our agents

Standard API prices

We run the agents for you and charge only what the models cost.

Early access

We’re taking on a few product teams now. Two teams outside Zarvo already use Relay on real work.

You get

  • We set Relay up with your team and run your first project with you.
  • Your feedback shapes what we build into Relay next.

You need

  • A GitHub repository the agents can open pull requests on.
  • Nothing else up front.