Tuesday. A developer with many repos.
Many projects, each with its own rules.
Your startup ships every day. The day job's repos want care, not surprises. The library you maintain has forty open issues and a weekend.
Without one place to run them
Every project needs a different leash, so in practice none of them gets an agent at all: one agent on one repo, in one terminal, while the rest wait for a free evening that doesn't come.
Wire it up once.
One behaviour at a time, each in the real dashboard.
-
Each repo goes as far as you say.
Build and open PRs on the startup, draft PRs only on the day job, plan and warn on the library, posting nothing.
How it works: Overhand reads a project's rule before it starts any agent there.
Your startupThe day jobThe library -
Each repo on its own agent and account.
The startup on your subscription, the day job on its company account, the library on a local model that costs nothing. Agents you're already signed in to are found, not pasted.
How it works: each project maps to an account; the login stays with the agent.
- Your startupsubClaude Code, on your subscription
- The day jobworkCodex, on the company's account
- The library$0OpenCode, on a local model
-
Review the night from one screen.
Every agent works on its own branch and stops at a pull request, with its checks and a preview link, waiting for you.
How it works: one agent per ticket, each in its own worktree, under one daily spend cap.
The brief
Your morning.
Overnight, three agents opened four pull requests across four projects. One production error broke at 00:58 and its fix is ready. One question waits, with a default.
- 4projects, four sets of rules
- 4pull requests to review
- $17of model spend, under the cap you set