Security at Punkto
Technical overview for DPOs, security teams, and anyone who wants to understand exactly how Punkto handles meeting data. No marketing copy — just the architecture.
The audio promise
Audio is never persisted to the database. It is processed in memory during transcription (a few minutes at most) then destroyed. The transcript and the AI summary are encrypted with AES-256 in a European database. The full chain, infrastructure included, lives in Europe. No provider subject to the US Cloud Act has access to your audio. If a failure interrupts processing, the raw audio is preserved on disk for at most 24 hours so it can be recovered, then automatically purged.
Encryption
Audio & transcript handling
Infrastructure
Access controls
Compliance
Zero audio retention — how it works
The audio pipeline is designed so that audio cannot be persisted, by architecture:
- Audio arrives as a multipart upload, held in a Node.js memory buffer.
- The buffer is streamed to the speech-to-text API over TLS.
- Transcript text is returned. The buffer is dereferenced.
- The Node.js garbage collector reclaims the memory.
- Transcript and AI summary are written to the encrypted database. The audio is not.
There is no fs.writeFile, no s3.putObject, no database column for audio path (it is permanently NULL). This is verifiable in the codebase.
FAQ
Questions or security reports
Security disclosures and DPA requests go to contact@punkto.app. We respond to security reports within 24 hours.