Login Book a demo

Security and reliability

Flagging risky agent behaviour shouldn't create new risk

Never in your agents' path. Sensitive data redacted at the door. Evidence your security team and your customers can check.

Book a demo
Settings / Security Your workspace

Security overview

Runtime posture for Larkspur Health

  • Access

    Scoped, read-only, revocable keys

    Scoped

  • Sensitive data

    PII and PHI redacted at ingestion

    On

  • Agent latency added

    Never in the request path

    0 ms

  • Evidence log

    Append-only, hash-chained

    Verified
DeploymentSample

  • Hosting

    Ollive cloud, your cloud or on-prem

    On-prem

  • Tenancy

    Dedicated instance, no shared data

    Single-tenant

  • Data residency

    Conversations and evidence stay in your region

    Your region

  • Leaves your environment

    Alerts to Slack, Linear, email

    Alerts only

Deployment

Runs where your agents run.

  • Ollive cloud, your own cloud, or fully on-prem.

  • Single-tenant isolation for regulated teams.

  • Conversations and evidence stay in your region.

Access

Read-only. Never in the path.

  • Works with Langfuse, Braintrust, LangSmith, Arize and your own stack.

  • Scoped keys you can revoke at any time. No write access to anything.

  • Zero added latency. Ollive never touches a live request, so every reply runs at full speed.

Connected sourcesSample

  • Langfuse prod-voice

    scope: traces:read, rotated 12 days ago

    Read-only

  • Braintrust servicing-agent

    scope: logs:read, rotated 3 days ago

    Read-only

  • S3 call-transcripts

    role: ollive-reader, prefix-limited

    Read-only

  • Write access requested

    never — Ollive has no write permissions

    None
Ingestion boundaryRedaction on

Record as it leaves your store

"Patient Maria Lopez, DOB 04/11/1986, member ID HX-77310 asked to reschedule…"

What Ollive stores

"Patient [NAME], DOB [DATE], member ID [ID] asked to reschedule…"

  • Retention: voice transcripts

    45 days

  • Retention: liability evidence

    Policy term + 1 year

  • Encryption

    TLS 1.2+ and AES-256 at rest

Data protection

Sensitive data stops at the door.

  • Names, IDs and card numbers are masked before anything is stored.

  • Retention you set, per agent.

  • Never used for training. BAA signed before ingestion.

Compliance

We hold ourselves
to the rules we check.

  • SOC 2 Type II

    Independent audit of our controls.

    In progress

  • HIPAA

    BAA available for healthcare agents.

    In progress

  • Penetration testing

    Third-party penetration test.

    Q4 done

  • Frameworks

    NIST AI RMF, EU AI Act, Colorado SB 24-205.

    Mapped

Frequently
Asked Questions

Take on bigger AI deployments
without taking on unlimited AI risk.

Book a demo