Core principles
-
Your data is never used to train models. Not ours, not third-party. This is contractual, not just policy. Your conversations teach your brain only.
-
We cannot access your content. We don't read your transcripts, listen to recordings, or review extracted insights. Infrastructure access is limited to encrypted metadata for debugging.
-
Deletion is real. When you delete data or cancel your account, it's gone. Not soft-deleted. Not archived. Permanently removed from all storage and backups within 30 days.
-
You own everything. Your transcripts, insights, knowledge graph, company state - all of it. Export anytime in standard formats. No vendor lock-in by design.
What we collect
Audio
Captured on your device, streamed for transcription
never stored by us
Transcripts
Text output from transcription, stored encrypted
Insights
Decisions, tasks, topics extracted by AI
Graph
Relationships between people, projects, decisions
State
Synthesized company understanding (themes, tensions)
Metadata
Timestamps, recording duration, seat usage
We do not collect: email content, browser activity, screen recordings, file contents, or anything outside of conversations you explicitly record with ambient.
Third-party processors
We use specialized services for specific tasks. Each processes only the minimum data required and is contractually bound to not retain or train on your data.
| Service |
Purpose |
Data shared |
Retention |
| Deepgram |
Speech-to-text transcription |
Audio stream (real-time) |
Not retained |
| Anthropic (Claude) |
Extraction & synthesis |
Transcript text |
Not retained |
| OpenAI |
Embedding generation |
Text segments |
Not retained |
| Supabase |
Database hosting |
Encrypted application data |
Until deletion |
| Neo4j Aura |
Knowledge graph hosting |
Graph relationships |
Until deletion |
On Private Cloud and Enterprise tiers, you can eliminate third-party processors entirely by running transcription, extraction, and storage on your own infrastructure.
Deployment & data residency
Where your data lives depends on your deployment tier.
Audio leaves device
For transcription only
To your infra only
Never
Transcripts stored
Our managed DB
Your dedicated DB
Your machine
AI processing
Third-party APIs
Your API keys
Local models
Knowledge graph
Managed Neo4j
Your Neo4j instance
Local graph DB
Region control
US/EU
Your choice
Your premises
Data encrypted
At rest + transit
At rest + transit
Your encryption
Your controls
You can at any time:
-
Export all data - transcripts, insights, knowledge graph, company state. Standard JSON/CSV formats.
-
Delete individual recordings - removes transcript, all extracted insights, and graph relationships derived from it.
-
Pause recording - ambient only captures when you choose. No always-on listening.
-
Remove team members - revokes access immediately. Their contributions to the brain remain unless explicitly deleted.
-
Cancel and delete everything - 30-day grace period for export, then permanent deletion across all systems.
Compliance trajectory
We are building toward formal compliance certifications. Current status:
-
SOC 2 Type II - in progress. Expected completion with GA launch.
-
GDPR-ready - data minimization, right to deletion, data portability, and processor agreements in place.
-
HIPAA - planned for Enterprise tier with fully local deployment.