Part of our work comes from public procurement. Prompts with client data do not get to wander into a US cloud by accident. That is a hard requirement, not a preference.
Soriku is the routing layer. Local models via Ollama for privacy-sensitive work. External models via your own API key when the job is heavier. Applications send a prompt and get a structured answer back. They do not pick the model themselves.
We do not train models in Soriku. We do not keep prompt data beyond the session unless you turn logging on. Akyru will sit underneath as our own model when it is ready. Until then the rule stays simple: the prompt stays home unless you explicitly send it out.