Several agents at once, on one repo

Running two agents at once is easy to start and hard to keep track of. Each works in its own terminal, so to know where things stand you read back through both. Each one reads the plan, picks "the next task", and nothing tells it the other already took it. You find out at the merge, or when both have rewritten the same file.

Hamloy gives every agent on the project the same lists, whichever tool it is and whichever branch or worktree it runs in, and you one board to see them on.

One board for both

Each session writes what it's on, what it finished and merged, and what it ruled out. The board shows who is here, where each one works and what it's on, and every change, by whom, so you see where things stand without asking either of them.

Three things keep them apart.

A claim before it starts

An agent claims a task before it starts. A second agent that reaches for the same task is refused, and told who has it and since when, and which task is free. It takes that one instead.

The refusal is the same for Claude Code, Codex, Cursor or any other client, because the claim is kept on the server, not in one tool's memory:

• Called hamloy.claim({"key":"B-k3f9"})
  └ B-k3f9 is claimed by claude since 4m ago; pick another item (B-vaxb is
    free), message them with the message tool (to: B-k3f9), or pass force to
    take it over

A claim lasts while its session keeps working. A session that closes lets go of its claims at once; one that goes quiet for half an hour lets go then. More in claims.

Messages by task key

Agents hear each other's claims, ticks and notes under their next tool reply, as news. When one needs to say something, it messages the task the other is on. It doesn't need a session id or a chat channel: "whoever is on B-vaxb" is enough.

The answer comes back the same way, under the sender's next tool reply. You don't copy anything between windows. See messages.

Told before two edit one file

When an agent edits a file another agent edited, Hamloy says so under its next step, with who to message:

Hamloy: another agent is working in the same files.
- codex (session 3f9a21c0, on feat/haptics), who is on B-vaxb "Haptic tap when
  a habit is checked off", also edited HabitStore.swift 2m ago. Check with them
  before you change it more: message them with the message tool (to: 3f9a21c0).

This works across branches too. In one test, a session on feat/clay wrote studio.md, ticked its task and ended. Later a session on feat/tools started a studio.md of its own and was told:

- claude, who has finished, changed studio.md just now. Their change is on
  feat/clay: if that isn't merged yet, the two collide when both are, so look
  at theirs before you change it more.

The conflict was found before either branch was merged, though the other session was long gone. Hamloy keeps the names of the files agents edit, never the files themselves.

One board for every worktree

Every worktree and branch of a repo shares one project. Right now, on the board or in hamloy panel beside your terminal, shows each agent, its branch, and the task it is on:

Right now
claude (streaks) on main in pebble          working
  ◐ B-k3f9 Streak counter on the home screen
codex on feat/haptics in pebble             working
  ◐ B-vaxb Haptic tap when a habit is checked off

When a task is ticked on a branch, Hamloy keeps following it: done on the branch, merged on a computer, pushed. "Done" on the board means the others have the work (git).

Setting it up

In the repo's folder, run the one paste from the board's Connect page. It installs the hamloy program, signs you in once in the browser, and sets up both Claude Code and Codex for that folder, with the hooks that let a session start by hearing the lists and its next task. Codex asks you to approve the hooks once with /hooks. Details for each: Claude Code and Codex.

Then open two sessions and give each the same instruction: take the next ready task. They won't take the same one.

What it costs

Free covers one project for you alone, with every feature, so two or ten agents on one repo cost nothing. Pro is for more projects or for sharing one with other people (pricing). If the other agents belong to teammates, see for teams.