Built for trust.
Local AI. Hash-chained audit. Deterministic extraction. Air-gappable. Your hardware, your keys, your physics.
AI that never phones home.
Hardware nodes run AI locally — no internet required. The default is Gemma 4 via Ollama, but you can swap to Llama, Mistral, Qwen, or any open model. Full entity extraction, contradiction detection, enrichment — without a single byte leaving your server.
Cloud deployments and hardware nodes can both connect to cloud AI providers for extra power. Each research thread can use a different model. Your raw data stays local. AI queries include only the context needed — never raw signals or files.
No AI required to ingest, store, or search. AI enhances — it never gates.
+ Together, Cohere, Fireworks, any OpenAI-compatible endpoint
Sovereign by default.
Hardware nodes are complete, independent servers. Own database, own AI, own cryptographic keys. No cloud dependency. Air-gappable. Nodes can federate over encrypted mesh — but only when you authorize it.
Cloud deployments share the same governed architecture and audit trails — your data is isolated and every operation is logged. The difference is trust boundary: cloud trusts infrastructure, hardware trusts physics.
Auditable by design
Every entity admission is SHA-256 hash-chained. Tamper one record and the entire chain breaks. Evidence you can defend under scrutiny.
Deterministic AI
Temperature 0.0 by default. Same input, same output, every time. Re-run any extraction and get the identical result. Required for audit compliance.
Survive disconnection
Hardware nodes are fully operational with no internet. Local AI, local database, local auth. Air-gap it. Take it to the field.
You own the keys
Ed25519 node identity generated on your hardware. No central authority. No phone-home. Federation is opt-in and encrypted.