Sovereign AI
Your commercial data is the last thing that should train someone else's model
Capra's intelligence runs on open-weight models deployed inside your own infrastructure. Pipeline notes, customer emails, prices and purchasing history stay where you put them.
Models on your hardware
Open-weight inference deployed to your machines. Pick the chat and reasoning model per orbit; Capra tunes context and token budgets to it.
Strict orbit isolation
Every table is scoped to the orbit and enforced at the data layer, so one orbit can never read another's records, documents or stock.
Encrypted mail credentials
SMTP and IMAP secrets are encrypted with AES-256-GCM and only ever decrypted server-side when a message is sent or fetched.
Roles & permissions
Operator roles, seat-level access, scoped invitations and permission checks on every server call — not just hidden buttons.
Audit & retention
Actions are logged with actor, object and time, with retention windows you configure per orbit.
Signed integrations
Webhooks are signed and idempotent; API keys are orbit-scoped and revocable at any moment.
What we never do
Three commitments, in plain language
No third-party AI vendor
Your records are not sent to an external model provider to be processed or retained.
No hidden data exhaust
We don't mine your pipeline to build features for someone else's product.
No lock-in on your content
Contacts, deals, documents and stock export cleanly whenever you ask.
Want the technical detail?
We'll walk your security team through deployment topology, model hosting, isolation and audit.