A fleet you own
Agents run continuously on hardware you already have. No metered token budget that runs out after a few hours — your binding constraint is compute, and compute is abundant.
Your own AI coding team,
running 24/7
Keikaku is a local-first agentic coding platform. You give it outcomes; it plans, implements, and reviews — using AI agents you host yourself, accented by frontier models when you choose.
Self-hosted or cloud — one codebase, your choice.
One outcome in. Plan → implement → review out.
The full loop, automated
Hand it an outcome. Keikaku plans the work, sizes it into right-fit tasks, implements with least-privilege per task, then verifies and reviews — looping until it's done.
Frontier models, first-class
Plug Claude and other frontier models in as agents for the hard parts. Every executor — local worker, downloaded agent, or frontier model — speaks one orchestration contract.
Self-healing by design
If a human has to diagnose and re-word a task for the agent, the product has failed. Recovery is built in — auto-repair, full-file recovery, and an inline fix loop.
Inverted scarcity
Hosted assistant
Binding constraint is token budget — finite, metered, gone after a few hours.
Keikaku
Binding constraint is per-token model capability — but compute and time are abundant. Spend cheap local compute to compensate: tight context, right-sized tasks, background self-maintenance.