Launching today
DeltaMemory
Fastest cognitive memory for AI Agents
47 followers
Fastest cognitive memory for AI Agents
47 followers
AI agents are getting smarter, but they still forget everything between sessions. We built DeltaMemory because we kept hitting that wall. Not a vector DB, not RAG, a real memory layer that extracts facts, builds a knowledge graph, and actually learns over time. Rust native. #1 on LoCoMo. 2x faster than Mem0 and other alternatives. 97% cheaper at scale.
DeltaMemory
Hey Product Hunt
We kept building AI agents that impressed people in demos and frustrated them in production. Not because the LLM was bad, but because every session started from zero. No memory of past conversations, preferences, or context. Just a smart tool with amnesia.
We tried vector DBs. We tried RAG. They retrieve similar chunks, they don't actually *remember*. There's a difference between finding a document and understanding a relationship built over 50 conversations.
So I started over, in Rust, and built the memory layer I wished existed. Not a wrapper. A full cognitive engine, knowledge graphs, biological forgetting, structured fact extraction, hybrid retrieval. The kind of memory system that actually makes agents feel like they know you.
The approach evolved a lot. Early versions were too complex to integrate. We stripped it down to three core actions: ingest, recall, reflect. Turns out that's all you need, if the engine underneath is doing the hard work properly.
We're #1 on LoCoMo benchmarks. 2x faster than Mem0 and other alternatives.
Happy to answer anything - use cases, or why Rust. Ask away 👇
DeltaMemory
@vouchy Hey Van, It wasn’t one big failure. It was repetition. After 50+ conversations, it still forgot stable facts and asked the same questions again. That’s when it clicked, RAG retrieves similar text and memory preserves state. We didn’t need better top-k. We needed structured and persistent memory.