Calls are full of things that shouldn’t live on someone else’s server: pricing you haven’t published, a prospect’s candid assessment of their current vendor, a customer’s account details, the internal aside they shared because they trusted the person on the call.
Most AI call assistants record that audio, store it, and process it later. Maax doesn’t. Your audio is transcribed as it streams and immediately discarded. Not stored temporarily, not stored encrypted — never stored. And unlike almost every tool in the category, Maax doesn’t store the transcript either.
Stored audio is a liability, not a feature
An audio recording is about the most sensitive artifact a call tool can hold. It contains voices — biometric identifiers — plus everything that was said and who said it. Stored at scale, it becomes a breach target, a compliance burden, and the thing a prospect’s security team asks uncomfortable questions about.
Voice recordings also sit in the heaviest regulatory category: the GDPR treats them as personal data with strict consent and protection requirements, and recording rules vary by jurisdiction. Most of that complexity simply evaporates when there is no recording.
How the architecture works
Captured on your own computer. Maax doesn’t send a bot into your call. The desktop app captures the system audio and your microphone locally — nothing joins the call or announces itself to the other side.
Transcribed on the fly. The audio streams to the speech recogniser and comes back as text, word by word. The audio exists only in transit, for the moment needed to transcribe it.
Analyzed in the moment. The live transcript is sent, chunk by chunk, to detect questions and hard moments and to answer them from your agent. Each request is processed and forgotten.
Ephemeral sessions. The transcript lives in your window while the session runs. Press Stop and it’s gone. Nothing about a live session is stored — no transcript, no answers, no history.
What is kept is deliberately small: your account, the knowledge you gave your agents, the gaps — questions your agent couldn’t answer, saved so you can close them — and a ledger of the credits each request used, never its content.
Control over what is stored
- Your knowledge, your call. Edit or delete any Q&A pair, page or file at any time; delete an agent and its search index goes with it.
- Gaps until you say so. Answer a gap and it becomes a pair; delete it and it’s gone.
- Delete the account, delete everything. Account deletion purges agents, sources, indexes, pairs, gaps and the ledger.
- No model training. Your content and your calls are never used to train AI models — ours or our providers’.
The boring-but-critical layer
The infrastructure underneath is what a security reviewer will actually ask about: data encrypted at rest and in transit across the entire pipeline, providers holding SOC 2 Type II attestations, sign-in through a dedicated identity provider, and a Data Processing Agreement that names every sub-processor and what each one receives. The privacy policy spells out the rest.
Privacy as a sales advantage
Here’s the practical upside: when a prospect asks “is this call being recorded?”, your answer is no. Nothing joined the call, no recording exists, and no transcript will exist ten minutes from now. That’s a fundamentally easier conversation than explaining someone else’s audio retention policy.
Get the answers. Keep nothing you don’t need.
See the architecture in action. Start free — live answers on your next call, with zero audio stored.