Offline & air-gapped
Some code can't go near a cloud AI — classified work, regulated industries, isolated networks, or plain caution. Keikaku runs entirely on your own hardware, with open models, so agentic coding works even with the network cable pulled.
What it is
A deployment where the platform, the models, and your code all live inside your perimeter. No calls to a frontier API, no telemetry, no source leaving the building. The same plan/implement/review workflow — just with nothing on the outside.
How it works in Keikaku
- Self-hosted platform. The whole server (API + app) runs as a single container on your infrastructure — no dependency on our cloud.
- Local models. Agents run open coding models through Ollama on your own GPUs. Benchmark your hardware to pick a model, pull it once, and you're independent of any external inference.
- Open-weights parity. Every capability — planning, implementing, reviewing, answering queries, brand extraction — is built for the open/self-hosted agent path. Nothing is gated behind a frontier model, so an air-gapped install is fully functional, not a stripped-down mode.
- Secrets stay local. An encrypted, write-only secrets vault lives in your own database; agents fetch credentials over your network only, audited.
- Licensed offline. Self-hosted entitlements verify from a signed license key offline — no phone-home required.
Where it shines
Defence and government, healthcare and finance, on-prem enterprises, and anyone whose security posture says source code and prompts must never touch a third-party model. If it has to run inside the fence, it can.