We've had a lot of enterprise security reviews ask the same questions: what do you store, for how long, and can you turn it off? This post answers all of that. We're also HIPAA certified and SOC 2 Type 2 certified, the report is linked at the bottom.

What Anam stores by default

By default, Anam retains the following for 30 days:

• Session recordings: user audio and persona video. Can be disabled on a per session basis via the SDK.

• Conversation transcripts: STT output and LLM responses. Automatically deleted after 30 days.

We also retain session metadata permanently: session ID, timestamps, duration, org ID, persona ID, and geo data. This contains no conversational content. It's operational data for billing and debugging.

One thing worth noting: Anam never receives the end-user's camera feed. Inbound video is not part of the pipeline.

Zero data retention

ZDR is available as an enterprise feature. When enabled, no session data (recordings, transcripts, or conversation logs) is stored after a session ends. What remains is only what's needed to run the service: session timestamps, usage metering, error codes, performance metrics, and infrastructure logs (retained for 30 days).

Enable it via Anam Lab (Build → Advanced tab → persona settings) or via the API:

zeroDataRetention: true

Full configuration docs at anam.ai/docs/concepts/privacy.

Two things ZDR doesn't cover: images used in persona creation and files uploaded to knowledge bases remain stored regardless of ZDR settings. Support and troubleshooting capabilities are also reduced under ZDR, since we have less data to work with.

How different integration modes affect what we see

The integration mode you choose determines what data Anam processes in the first place. In Custom LLM mode, we handle STT but the LLM runs on your side. We never see the transcript. In Custom LLM+STT mode, we only receive text. In LiveKit, Pipecat, Agora, and audio passthrough modes, Anam handles face generation only and receives no conversation content at all.

If minimising data exposure is a priority, ZDR on top of a LiveKit or audio passthrough setup means Anam processes nothing conversational and stores nothing.

Third-party providers

Cartesia and ElevenLabs both operate under zero data retention agreements with us. Text sent for TTS synthesis is processed and not stored. For LLMs in Turnkey mode: Azure OpenAI has training and logging disabled, Groq operates under a ZDR agreement, and Gemini runs under standard terms.

STT (our self-hosted Deepgram model) and face generation (Cara) run on Anam's own infrastructure. No third-party exposure on either.

Trust and compliance documents

Everything below is available on our trust portal:

For questions on data handling, ZDR configuration, or HIPAA requirements, contact your Anam account representative.