Guardrail
PHI Rail
Protected health information passes through a layered redaction and pseudonymization pipeline before it can reach any non-BAA destination. Real PHI never leaves your boundary un-scrubbed.
Security & compliance
HIPAA-safe by default. Isolated by design.
Ambrose handles PHI-adjacent work every day, so safety isn’t a setting — it’s the default. Here’s how your data stays yours.
Guardrail
Per-agency isolation
Every record carries an agency id, enforced by application filters and database row-level security. One agency can never see another’s keys, chats, credits, or contacts.
Guardrail
You hold the keys
Run on Amazon Bedrock, or bring your own Anthropic key. Provider credentials are resolved per-agency and never exposed to the model.
Guardrail
Full audit trail
Every AI call and every staged action is logged and attributable. Sensitive actions are previewed and require your approval before they run.
Guardrail
No cross-tenant prompts
Skills, files, and conversations are scoped to your agency. Nothing from another tenant is ever injected into your agents’ context.
Guardrail
Least privilege
Agents get only the tools and integrations you attach. The in-app assistant can create and update, but never deletes.
The short version
Ambrose is built so an agency can safely put AI to work on real client data. Protected health information is redacted and pseudonymized before it can reach any destination that isn’t covered by a Business Associate Agreement. Each agency runs in its own isolated tenant, with its own keys and data. And because every action is logged — and sensitive ones require your approval — you always know what your agents did and why.
Have a security or compliance question? Get in touch and we’ll walk you through it.
Ready when you are.
Start free — your keys, your data, HIPAA-safe from the first message.
Start free