Security

How the environment is set up

Written for the people who have to sign off on it.

Single-tenant deployment: Runs on AWS, on-prem or dedicated cloud. No shared tenant.
Encryption: Encrypted in transit and at rest.
Zero data retention on model calls: Model calls are ephemeral. No copy is retained by the model provider.
Client Codex and history stay put: The Codex and historical memos remain in the client environment.
No IP harvesting: LogiCov does not harvest underwriting IP to resell.

Assurance: isolation by design, running on AWS infrastructure, with program evidence produced in your production environment.

Chinese walls are structural. One client's files, policy, and memo language do not enter another client's tenant.

Containment

Same task. Two very different paths.

Containment, not a smarter model.

Drop it in a model

One prompt carries the file, the policy, the industry question, and the memo ask. The file leaves the workflow boundary. The model can answer outside the question. There is no separate check before the draft exists.

One window. No harness. No owner.

Narrow agents, one tenant

The file stays inside the institution's environment. A financial agent only spreads. An industry agent only reads the sector. A policy agent only checks their rule book. A guardian agent reviews before a human signs. Nothing is pooled. The analyst still submits to committee.

Each agent does one job. The institution owns the conclusion.

A general model is built for every question. This is built for one file: their policy, their template, their tenant. The draft does not wander into a chat.

LogiCov harness

File in
Narrow agents
Spread, sector, policy
Guardian check
Kill switch. A reviewer can stop the draft before it exists. Nothing leaves the tenant.
Human sign-off
Committee

Ingestion and synthesis. The institution owns the credit conclusion.

Security questions or a diligence questionnaire? kris@logicov.com