Local-first personal context layer for AI agents.
Your apps and AI assistants should know your preferences without you explaining them repeatedly. Personal Layer observes your digital behavior, builds a private profile on your device, and shares scoped context with the apps you trust.
Production Status: This codebase is transitioning from prototype to production. The Python FastAPI backend is the local reference runtime, and the iOS app is the native production target.
Connected apps and integrations
|
v
Local ingest gate (blocks credentials/secrets only)
|
v
Encrypted raw vault (AES-256-GCM, device key)
|
v
Persona synthesis (Core ML all-MiniLM-L6-v2 + STEP-BACK profiling)
|
v
Egress privacy filter (PII scrubbed, raw content stripped)
|
v
Scoped context bundles → MCP / SDK / Extension / Assistant
- Raw data stays local in an encrypted vault.
- Thin cloud (Supabase) stores only developer registry, consent metadata, and push routing.
- Every outbound path is filtered before it leaves your device.
cd backend
pip install -r requirements.txt
python main.py
# Dashboard: http://localhost:7823scripts/build-ios.sh| Path | Purpose |
|---|---|
backend/ |
Python FastAPI reference runtime |
native/ios/ |
Swift app + GRDB |
extension/ |
Chrome MV3 extension |
sdk/python/ |
Python SDK |
sdk/javascript/ |
JavaScript/TypeScript SDK |
supabase/ |
Thin cloud migrations + edge functions |
tests/ |
Production hardening tests |
docs/ |
Architecture, privacy, threat model |
- AES-256-GCM encrypted local vault
- OS keychain / Secure Enclave for device key and OAuth tokens
- Egress filter on every outbound path (MCP, API, SDK, extension, cloud, notifications)
- CORS restricted to localhost + extension origins
- Row-level security on all cloud tables
- Silent APNs pushes only (no behavioral text in payload)
- Security reports: security@personallayer.dev
MIT