Your meetings are yours.
How Briva handles the data your team trusts us with — what we encrypt, what our vendors can do with it, how to delete everything, and what we're working towards next.
Last updated: 11 May 2026
Deepgram and Anthropic do not train models on data we send via their APIs.
TLS 1.2+ in transit, AES-256 at rest. Industry standard, by default.
Delete any meeting anytime. Configure audio auto-delete per workspace.
Where your data goes
When you record or upload a meeting, the audio flows through these systems and nowhere else:
- Your browser or desktop app → Briva over TLS. The desktop app uses the same HTTPS endpoints — there's no separate channel.
- Briva → Deepgram for transcription. Audio is sent over TLS; Deepgram returns text. Deepgram is a US-based, SOC 2 Type II certified provider and does not train models on customer audio.
- Briva → Anthropic for AI summary and chat. Transcript text is sent over TLS; Claude returns the summary. Anthropic does not train models on data submitted through the API by default.
- Briva → Voyage AI for cross-meeting search embeddings. Short chunks of transcript text are converted to vectors stored in our database. Voyage does not retain or train on submitted content.
- Storage in Supabase (Singapore region) — audio, transcripts, summaries, embeddings. Supabase uses AWS infrastructure with SOC 2 Type II certification and AES-256 encryption at rest.
We do not sell data. We do not share data with third parties for advertising or analytics purposes. Vendor list above is the complete pipeline.
Encryption
- In transit: TLS 1.2+ on every request — including the desktop app, the web app, and all vendor calls.
- At rest: AES-256 (database + object storage) via Supabase / AWS.
- Secrets: all third-party API keys (Deepgram, Anthropic, Voyage, Resend, Stripe) are server-side only — they never reach the browser. Live transcription uses short-lived scoped tokens issued by our server.
- Backups: Supabase performs daily encrypted backups with point-in-time recovery for our database tier.
Retention & deletion
You choose how long Briva keeps your raw audio:
- Keep forever (default) — audio stays available for re-summarisation.
- Delete after processing — audio is deleted as soon as transcription + summary complete. Transcript and summary are retained.
- Delete after 7 / 30 days — a scheduled job removes the audio file on that timeline. Transcript and summary are retained.
Workspace owners configure this in Settings → Team → Audio retention. Individual meetings can also be deleted on demand — audio, transcript, summary, embeddings, comments, and tasks are removed permanently within minutes.
Account deletion is honoured within 30 days; all personal data is purged from primary storage. Anonymised billing records are retained where required by HK / EU tax law.
Workspace isolation
Every meeting belongs to exactly one workspace. Access is enforced at the database layer using PostgreSQL row-level security (RLS) — there is no application-layer-only check.
- Members: only users in a workspace's member list can read its meetings.
- Removed members: losing membership instantly revokes access — their session can still issue requests, but RLS denies them.
- Roles: owner / admin / member with explicit write permissions per role.
- Share links: non-guessable IDs (random tokens, never sequential), optional password protection, revocable from the share dialog.
Recording consent
Recording laws vary by jurisdiction — Hong Kong and most APAC markets allow one-party consent; California and the EU require all-party consent in many contexts. Briva surfaces a reminder before any live recording starts so the person recording is responsible for verifying consent with the other participants.
Our commitment: we do not auto-join meetings on a user's behalf without their explicit action. Every recording is initiated by a Briva user who is present in the meeting or has explicit recording authority.
Our vendors
| Vendor | Purpose | Region / certs |
|---|---|---|
| Deepgram | Speech-to-text | US · SOC 2 Type II · No training on customer audio |
| Anthropic | AI summary & chat (Claude) | US · SOC 2 Type II · No training on API data by default |
| Voyage AI | Embeddings for cross-meeting search | US · No training, no retention beyond request |
| Supabase | Database + audio storage | Singapore · SOC 2 Type II · AES-256 at rest |
| Vercel | Application hosting / edge | Global · SOC 2 Type II |
| Resend | Transactional email (recaps, invites) | US · GDPR-compliant |
| Stripe | Subscription billing | US · PCI DSS Level 1 · No payment data on our servers |
Compliance roadmap
We're a young company and we're upfront about what we have versus what we're working on:
- Today: All vendors above are SOC 2 Type II or equivalent. Our own application enforces RLS, TLS, AES-256, and least-privilege secrets.
- Q3 2026: Third-party penetration test against the production stack. Findings published as a redacted summary on request.
- Q4 2026: Data Processing Addendum (DPA) template for enterprise customers; GDPR alignment for EU prospects.
- 2027: SOC 2 Type II audit. Existing controls are designed to map cleanly; the audit formalises them.
- Later: ISO 27001. We'll prioritise it once paying enterprise customers ask.
Reporting a security concern
If you believe you've found a vulnerability, please email sattarikram81@gmail.com with a description and reproduction steps. We'll acknowledge within one business day and follow up with a triage timeline.
We do not currently operate a paid bug bounty programme, but we're happy to publicly credit responsible disclosures.
Need a one-pager for your IT team?
We have a vendor-evaluation packet covering data flow, encryption, retention, and our compliance roadmap. Tell us a little about your team and we'll send it over.