Coming soon For developers

Bring your coding agent. Code Relay runs the review loop.

Code Relay is one more AI-assisted coding tool in your kit. You don't install or run it: you connect your own Codex or Claude to it over MCP. Code Relay hands your agent the task, reviews what it builds and sends back findings until the task is accepted. Then you carry on with your own flow.

  • Your Codex or Claude, connected over MCP
  • One task at a time, on any project
  • Never pushes, merges or deploys for you
Illustration. Your agent implements, Code Relay reviews, and the rounds continue until the task is accepted. Nothing is pushed; the result comes back to you.

How it works

How one task runs through Code Relay

You bring the task and your agent. Code Relay brings the handoff, the review and the record. You keep the result.

  1. Start from one task

    The unit of work is a single task, from whatever project you're on. Code Relay doesn't take a whole backlog at once; you might run two tasks side by side, or one after the other.

  2. Connect your agent over MCP

    Connect your own Codex or Claude session to Code Relay over MCP. There is nothing to install or host: your agent stays yours, and Code Relay is the other end of the conversation.

  3. Your agent gets the implementation handoff

    Code Relay gives your agent the handoff: what to build, the scope, and the acceptance criteria the work will be reviewed against. Your agent implements and reports back.

  4. Review returns findings

    The reviewer treats the report as claims and checks them against the acceptance criteria. Anything that falls short comes back to your agent as a finding to fix.

  5. Fix, review again, until accepted

    Rounds of implement, review and fix continue until the reviewer accepts the task. The implementer never accepts its own work, and every handoff, finding and verdict is kept as a durable record.

  6. Back to your own flow

    Code Relay doesn't push, merge or deploy for you. You get the accepted result back and decide what happens next, the same way you would with any other tool.

Review

Two ways to review, and you can use both

Either way, the implementer and the reviewer are separate conversations, and only the reviewer grants technical acceptance.

Code Relay reviews

Code Relay reviews what your agent implemented on its own side and returns findings for your agent to fix. You only connect the implementer.

Your second session reviews

Connect a second Codex or Claude session over MCP as the reviewer. It receives the review handoff: what to review, and against what.

Any project

It works wherever your tasks come from

Code Relay doesn't depend on any marketplace or company. If you have a task and an agent, you can use it.

A task you won on Forja

Use Code Relay on a Forja task if you want to. Forja developers are not required to use it.

A client of your own

A client you found directly, or through Upwork or anywhere else.

Your day job

A private project at the company you work for.

Principles

What Code Relay holds itself to

  • The reviewer decides

    The usual wayThe session that wrote the code also says it's done.

    In Code RelayImplementer and reviewer are separate conversations. Only the reviewer grants technical acceptance.

  • Findings, not claims

    The usual wayAn agent reports success, and that's the report.

    In Code RelayThe implementer's report is a set of claims. The reviewer checks them against the acceptance criteria and returns findings until they hold.

  • One task per job

    The usual wayOne long session tries to carry a whole feature.

    In Code RelayEach job is a single task with its own scope and acceptance. On Code Relay's side, each task gets its own isolated git worktree.

  • Durable records

    The usual wayThe task and its findings live in chat scrollback.

    In Code RelayHandoffs, findings and verdicts are durable records, so the work survives a session being replaced.

  • You decide what ships

    The usual wayAgents get write access to the remote.

    In Code RelayIt never pushes, merges or deploys for you. You get the result back and decide.

Fully autonomous mode

Forja plans to use Code Relay for single tasks

Code Relay can also run a task entirely on its own, with no human session connected over MCP. Forja plans to use this mode for individual tasks that go unassigned.

  • Autonomous

    Implement, review, fix, accept

    Code Relay's own implementer and reviewer work the task as separate conversations, in an isolated git worktree, until the reviewer accepts it.

  • Forja

    When a task gets no developer

    If a task gets no bids or the client leaves it unassigned, Forja plans to send it to an agent. Code Relay will handle individual tasks.

  • Either-or

    Single tasks, not batches

    Forja will route each task to one service. Code Relay works one task at a time. Ad Actum handles batches, such as a feature built from several interdependent tasks.

  • Disclosed

    Always shown as agent work

    Forja will tell the client whenever an agent did the work.

Status

Coming soon, and said plainly

Connecting your own agent to Code Relay isn't open yet. Here is where things stand.

  • Coming soon

    Connect your agent over MCP

    The developer offering, your own Codex or Claude connected to Code Relay, is not open yet.

  • No install

    Nothing to install

    Code Relay is not software you download or self-host. You connect to it; you don't run it.

  • Now

    Join the waitlist

    Leave your details and we'll write when connecting your agent opens. Signing up doesn't give you access today.

FAQ

Straight answers

What you connect, who reviews, and what Code Relay won't do.

Do I install or run Code Relay?

No. Code Relay isn't distributed, and there is nothing to install or self-host. You connect your own agent to it over MCP.

Which agents can I connect?

Codex or Claude, as the implementer and, if you want, as a second session for review.

Who reviews the work?

Code Relay can review on its own side and return findings, and you can also connect a second Codex or Claude session as the reviewer. Either way, the reviewer is a separate conversation from the implementer, and only the reviewer grants technical acceptance.

Can I use it today?

Not yet. The developer offering is coming soon. Join the waitlist to hear when it opens.

Do I need Forja?

No. Code Relay works on any project: a Forja task, a client you found yourself or on Upwork, or a private project at your company. Forja developers aren't required to use it either.

How big is one job?

One task. Code Relay can run tasks in parallel internally, but each job is a single task. You might run two at once, side by side or one after the other.

Does it push, merge or deploy?

No. It doesn't push, merge or deploy for you. You get the accepted result back and decide what to do with it.

Can it work with nobody connected?

Yes. In fully autonomous mode Code Relay implements, reviews, fixes and accepts a task on its own. Forja plans to use this mode for single tasks that go unassigned.

Is Code Relay part of Ad Actum?

Code Relay and Ad Actum are independent services. Forja plans to use one service per task.

Waitlist

Join the waitlist

For developers who want to connect their own Codex or Claude to Code Relay. Signing up doesn't give you access today; we'll write to you when it opens.