Astos

Before every call, an automated pipeline runs three checks in sequence: consent, do-not-call status, and the permitted calling window. A failed check cancels the call, so blocked calls never dial. Every outcome is written as an immutable record alongside a full audit log, and recording disclosures are applied per jurisdiction. The audit trail can be exported as CSV.

The voice loop runs end-to-end on ElevenLabs behind a dedicated Python service: speech-to-text, an LLM-driven conversation, text-to-speech, and the telephony connection. This replaced an earlier multi-vendor design. Audio and transcript are recorded and returned through authenticated webhooks, together with status, duration, intent, and sentiment. Because the voice API offers no per-call training opt-out, that requirement is bound contractually in the data processing agreement.

Access tokens are short-lived (15 minutes) and paired with rotated, hashed refresh tokens. Multi-factor authentication is available via authenticator app or email code, backed by login lockout and rate limiting. The API uses strict security headers and a locked CORS policy, and outbound webhooks are signed with HMAC-SHA256 and retried with backoff. Administrative tooling is excluded from production.

Yes. The platform is multi-tenant: every record carries an organisation id, and every query is scoped to that organisation from the JWT, so there is no cross-tenant access. Nothing is hard-deleted; the system uses soft deletes throughout. Audit and compliance logs are immutable and can never be modified, and the backend and the AI service each run their own separate database.