**Cortex - AI Memory System Built Entirely on Convex**
Hey Convex community! 
I've built Cortex - an open-source AI memory system for agents, powered entirely by Convex.
What it does:
Gives AI agents infinite context with intelligent memory:
-
99% token reduction via retrieval-based recall
-
ACID + Vector in ONE database (Convex's hybrid architecture FTW)
-
Hive Mode - Multiple AI tools share memory (Cursor + Claude + custom agents)
-
Fact extraction - 60-90% storage savings
-
Context Chains - Multi-agent workflow coordination
-
One-click GDPR deletion across all layers
Why Convex:
Tried Pinecone + Redis + custom logic = polyglot nightmare. Convex changed everything: ACID transactions + real-time reactivity + automatic indexing in one place. This genuinely wouldn't exist without Convex's architecture.
Origin:
Built for Project Constellation (multi-agent system for MS Teams). Needed agents that remember forever with semantic search. Existing solutions forced choosing Vector (no ACID) OR traditional DB (no search). Convex let me have both.
Ready now:
- TypeScript:
- Python:
- Edge runtime compatible, 579 tests, Apache 2.0
GitHub: https://github.com/SaintNick1214/Project-Cortex
Feedback wanted:
- API ergonomics (Convex patterns I'm using)
- Performance optimization ideas
- Real problems this solves for you?
Thanks for building an incredible platform. This would've been 10x harder anywhere else!
P.S. Applied to Convex for Startups 2 weeks ago - would love to be closer to the community and roadmap. Building memory systems on Convex feels like the future.
I've built Cortex - an open-source AI memory system for agents, powered entirely by Convex.
What it does:
Gives AI agents infinite context with intelligent memory:
-
99% token reduction via retrieval-based recall-
ACID + Vector in ONE database (Convex's hybrid architecture FTW)-
Hive Mode - Multiple AI tools share memory (Cursor + Claude + custom agents)-
Fact extraction - 60-90% storage savings-
Context Chains - Multi-agent workflow coordination-
One-click GDPR deletion across all layersWhy Convex:
Tried Pinecone + Redis + custom logic = polyglot nightmare. Convex changed everything: ACID transactions + real-time reactivity + automatic indexing in one place. This genuinely wouldn't exist without Convex's architecture.
Origin:
Built for Project Constellation (multi-agent system for MS Teams). Needed agents that remember forever with semantic search. Existing solutions forced choosing Vector (no ACID) OR traditional DB (no search). Convex let me have both.
Ready now:
- TypeScript:
npm create cortex-memoriesnpm create cortex-memories (5 min setup)- Python:
pip install cortex-memorypip install cortex-memory (full API parity!)- Edge runtime compatible, 579 tests, Apache 2.0
GitHub: https://github.com/SaintNick1214/Project-Cortex
Feedback wanted:
- API ergonomics (Convex patterns I'm using)
- Performance optimization ideas
- Real problems this solves for you?
Thanks for building an incredible platform. This would've been 10x harder anywhere else!
P.S. Applied to Convex for Startups 2 weeks ago - would love to be closer to the community and roadmap. Building memory systems on Convex feels like the future.
GitHub
A unified memory system that gives you everything in one package - production-ready memory that scales automatically, works with any LLM framework, supports any embedding provider, and requires zer...
