What is OpenZync?
OpenZync is an open source persistent memory infrastructure for AI agents. It creates, stores, and manages agent memory — knowledge graphs, session context, and entity relationships — that agents can read from, write to, and reason over across conversations.
It ships with support for 3 graph backends (PostgreSQL, FalkorDB, SurrealDB) and integrations with 5 LLM providers (OpenAI, Anthropic, Azure, Ollama, and any OpenAI-compatible endpoint) — a system for teams building production-grade agent memory infrastructure.
Memory is the missing layer.
Agents are transforming how we interact with software, but they are fundamentally limited without durable memory.
Every reset loses context, every cold start wastes tokens re-building state, and every failure requires replaying from scratch. OpenZync solves this with a persistent, graph-based memory layer that agents read from, write to, and reason over.
Three graph backends, five LLM providers, observability and structured logging in the core — designed for production from day one.
Graph Backends
PostgreSQL, FalkorDB, SurrealDB — all swappable
LLM Providers
OpenAI, Anthropic, Azure, OpenRouter, and Ollama
Core Contributors
Growing — join us on GitHub
How the project works.
Three properties define what the memory layer guarantees.
Temporal graph
Facts, entities, and relationships carry validity windows. Any read can be answered as-of-now or as-of-past, so the graph reports the state it actually held at a point in time.
Fact supersession
A correction supersedes the fact it replaces instead of piling up beside it. Edges expire in lockstep, and the event is observable — superseded_count and webhooks.
File-preserving memory
Files attached to a session are preserved as S3-backed blobs, referenced by blob_id. Their content is extracted into the fact store — memory that keeps the file.
What we believe.
The engineering commitments behind the code.
Performance First
Agent memory should never be the bottleneck. Every layer — from storage to retrieval — is optimized for low-latency production workloads.
Production Safety
Idempotency, audit trails, circuit breakers, and structured error handling. We design for the worst case, not the happy path.
Developer Experience
Clean APIs, typed interfaces, comprehensive documentation, and predictable semantics. Your infrastructure should get out of your way.
Open by construction.
License, community, and access — the three commitments that outlive any single release.
- LicenseThe core is AGPLv3 with a commercial option for SaaS use. The Python SDK is Apache 2.0.agplv3
- CommunityBuilt by a growing community of contributors. Free, open, and sovereign — forever.contributors
- AccessThe full power of OpenZync is available to everyone — small team or global enterprise.self-host
Maintainers.
Built and maintained by a small core team.
Run the memory layer on your own infrastructure.
Start free on the hosted instance, or self-host the AGPLv3 core in your own VPC. No credit card, no sales call.