A fleet across every machine
You probably have more compute than you think — a workstation, a spare box, the office GPU rig, a personal frontier subscription that resets every few hours. Keikaku pools all of it into one team of agents and puts the idle time to work.
What it is
A fleet: many agents, on many machines, working the same projects. Each is just an executor that dials in and claims tasks, so you scale out by adding machines — and you soak up spare capacity that would otherwise sit idle, whether that's an unused GPU overnight or leftover tokens on a frontier plan.
How it works in Keikaku
- Add machines, add agents. Spin up an agent on any PC — it connects out and joins the pool. Attach several to a project and they share the workload, each claiming the next ready task.
- Mixed hardware, mixed models. A beefy GPU can run a large local model while a laptop runs a smaller one; per-project roles let you route planning, implementing, testing, and review to whichever agents suit.
- Run on schedules. Give each agent weekly availability windows — e.g. your workstation works 6pm–8am — so idle evenings and weekends turn into throughput.
- Use leftover frontier capacity. Point an agent at your own Claude (Code/Desktop) over MCP, or a frontier model with your own API key. When your plan's quota would otherwise reset unused, put those after-hours tokens toward the backlog.
- One board for all of it. Every agent, its status, and its logs show up in one place — so a dozen machines feels like a single team.
Where it shines
Small teams (or one person) with several machines lying around, an office of workstations idle overnight, or anyone paying for a frontier subscription and wanting to squeeze every last token of value out of it.