System Overview
Learn about the Omni privacy-centric AI memory infrastructure.
Welcome to the Omni developer documentation. Omni is a developer-focused platform and open-source runtime designed to manage, secure, and sync agentic memory enclaves.
By shifting memory structures away from centralized multi-tenant databases and into isolated user enclaves, Omni guarantees that your users' private details are kept safe while providing standard OpenAI-compatible completions endpoints.
Core Pillars of the Architecture
Omni is built around three fundamental pillars:
1. Gateway API
A standard completions and memory ingest gateway compatible with typical LLM libraries. Easily ingest messages, extract structured facts, and request prompt-ready context blocks.
- Unified prompt compilation endpoints
- Mem0/Zep-style standard payload structure
- Zero overhead latency routing
2. Enclave Isolation
Instead of simple database row segregation, Omni automatically provisions physically separate Cloudflare R2 storage buckets and isolated Vectorize similarity search namespaces for every single workspace user.
- Absolute tenant physical separation
- Scalable, serverless on-demand provisioning
- High-efficiency query routing
3. Local-First Sync
Integrate local client databases (SQLite) running directly on user devices. Background synchronization passes coordinate encrypted coordinates and similarity parameters to keep plain text from ever leaving the local sandbox unnecessarily.
- Offline-first durability
- AES-GCM client-encrypted vaults
- Direct client-side similarity calculations
Next Steps
To begin provisioning your enclaves and calling standard routes:
- Check out the Quickstart Guide to test integration options with your active developer keys.
- Dive into Core Concepts to understand facts extraction, enclaves, and vectors.