Security · Sovereign edge
Run Seal in front of a local model
Decision-support on infrastructure you control. Seal clears the call on the box, signs ATTESTATION-v1 with a deployer-held key, and leaves a receipt a hostile reviewer can check offline. Foreign cloud is optional, not required.
What is hard to copy
Anyone can wrap Ollama. The product moment is operator-independence: stop the edge (or stop working with us) and every receipt already issued still verifies against the key you held, with no foreign provider in the check path.
- Checked before it acts, not a ledger written after
- Hostile handoff: pin the deployer key, flip one field, watch the signature die
- Same open ATTESTATION-v1 envelope and published verifier family as hosted Seal
- Counsel-facing handoff pack from the edge process, not a vendor dashboard login
What you get
- Ollama (or any OpenAI-compatible upstream) on your machine
- Seal sovereign-edge gateway on port 8080
GET /v1/handoff/{id}cover note + pin + tamper checklist- Smoke script that proves verify, issuer pin, and hostile fail in one run
Run it
cd sovereign-edge ./scripts/gen-key.sh docker compose up -d --build ./scripts/pull-model.sh ./scripts/smoke.sh
Source: the sovereign-edge directory of the Seal repository, shared with pilots on request. Chat: http://127.0.0.1:8080/v1 (Bearer sk-edge-demo).
Why this matters for sovereignty
Sovereign compute keeps data and inference inside your jurisdiction. That still leaves a gap: proving what the model decided in a single case, in a form that travels outside the systems that produced it. Seal is that missing half: a decision record that stays checkable wherever the question is asked.
Bring your model, your metal, your EU host. Seal sits in front and leaves a file a hostile reviewer can check when the cloud, or we, are no longer in the room.