Skip to content

Security & trust

Security, stated plainly

How Callario protects your workspace and your customers’ data today — access, isolation, AI safety, audit and integrations — and, by name, what is not built yet.

On this page

01At a glance

Controls that are built and running today.

Everything on this page is in the product now. What isn’t is listed separately, by name.

  • GDPR export & erasure
  • Hash-chained audit log
  • Admin-enforced 2FA
  • Five roles
  • No training on your data

Guardrails on every AI turn

Prompt-injection detection in English and Arabic, a grounding check and a toxicity block run on every turn; an injection attempt is blocked and handed to a person. Personal data is flagged and scrubbed from the traces we store.

An audit log you can verify

Configuration changes, publishes, takeovers, people and access changes, and API-key and webhook changes are hash-chained. Admins can verify the chain and export it as CSV. AI turns are kept in run traces.

GET /v1/audit/verify

Least-privilege access

Five roles — owner, admin, lead, agent, viewer — and two-factor authentication an admin can require for the whole workspace. API keys carry only the scopes you grant.

02Controls by area

What protects your workspace, area by area.

Each line is a control in the product today, checked against the code.

Access and identity

  • Five roles — owner, admin, lead, agent and viewer — and a roles sheet that says what each can do.
  • An admin can require two-factor authentication for the whole workspace, with set-up at sign-in and a seven-day grace period.
  • Recovery codes, an admin reset for a teammate who lost their authenticator, and every 2FA change on the audit log.
  • Sign out every other session in one click; password guessing is bounded per account.

Data protection

  • Tenant isolation in the database: row-level security keeps one workspace’s queries from reading another’s rows.
  • All traffic is encrypted with TLS. Application secrets and channel credentials are encrypted at rest in the database.
  • GDPR export and erasure for a customer’s personal data, reaching the AI and chat records as well as the core tables.
  • Consent on every customer record, and an AI knowledge card that shows exactly what the AI may know about them.

AI safety

  • Prompt-injection detection in English and Arabic, a grounding check and a toxicity block on every AI turn.
  • Personal data is flagged and scrubbed from the AI traces we store. It is not removed from the live reply — the agent needs it to help.
  • No customer content is used to train foundation models.
  • An eval gate blocks publishing an AI agent that fails a critical test question.

Audit and evidence

  • A hash-chained audit log of configuration changes, publishes, takeovers, people and access changes, and API-key and webhook changes.
  • Verify the chain from the console, and export the log as CSV with both hashes on every row, so an unfiltered export can be re-chained outside Callario.
  • The log reads as sentences — who changed what, from what, to what.
  • Every AI conversation keeps a run trace: what was retrieved, which checks ran, which tools were called.

API keys and webhooks

  • Scoped API keys — tickets:read, tickets:write, contacts:read — stored hashed, shown once, revocable at any time. To rotate a key, issue a new one and revoke the old.
  • Webhooks signed with HMAC-SHA256 over a timestamp and the raw body, in an x-callario-signature header.
  • Webhook secrets can be rotated; failed deliveries are retried with backoff and can be replayed.

03Not built yet

What Callario does not have today.

Listed by name so nobody has to infer it from silence.

  • SSO / SAML and SCIMThe code exists but is switched off. It is not available to workspaces yet.
  • Data residencyCallario runs in a single hosting region. There is no choice of region.
  • SOC 2Callario does not hold a SOC 2 attestation, and no audit has begun.
  • Uptime SLANo uptime commitment is offered during the alpha.
  • Volume-level encryption at restSecrets and credentials are encrypted in the database; full-volume encryption is not configured yet.

04Straight answers

The questions security reviews ask first.

Short answers. Design partners can go through the detail with us.

Do you train AI models on our data?
No. Customer content is never used to train foundation models. The AI processes your data only to answer, draft or classify for you inside Callario.
Where is our data hosted?
In a single hosting region; there is no choice of region today. Ask us for the current location through the contact page.
Can we sign in with SSO?
Not yet. SSO, SAML and SCIM are built but switched off. Today a workspace signs in with email and password, and an admin can require two-factor authentication for everyone.
Is personal data removed before the AI answers?
No — and we say so plainly. Personal data is detected and flagged on every turn and scrubbed from the traces we store, but the live reply keeps it, because the agent needs those details to help the customer.
Can we check the audit log ourselves?
Yes. An admin can verify the hash chain from the console and export the log as CSV with both hashes on every row, so an unfiltered export can be re-chained independently.
Do you have SOC 2?
No. Callario does not hold a SOC 2 attestation and no audit has begun. Design partners can review our current controls with us directly.

Design-partner programme · now onboarding

Bring your security questions.

We’ll show your team the controls, the audit log and the run traces in the live product.

Book a demoRead the DPA