News for hackers. Stuff that matters.

costinu

Re:Thoughts on memory layers like mem0 on Sunday November 15, 2026 @03:04PM (#50510162)
Local md files work fine when it's just you on your own machine, but for an app/service with thousands of users, using md files will be challenging.

You also need to handle duplicate or conflicting facts, and that becomes a burden as the memory grows.

Many app builders might start using a vector database. This will help with storage and organising data, but will fall short on keeping an accurate memory when facts change.

So I think there is a market, but only for the multi-user case.

Recalld is a memory layer for AI agents and chats. It can ingest conversations, documents or code. The input is decomposed into facts, then saved into a vector database. I built Recalld because I couldn't find a reliable memory layer that let my agents retrieve accurate data without noise.

The facts from each add operation are compared with existing facts, and Recalld then decides whether each one is an addition, an update or a replacement of existing facts. Agents can retrieve context via two methods. The first is search, which is faster and cheaper and performs a vector search. The second option is recall which uses an LLM to select relevant facts. In our published benchmark on LoCoMo benchmark run, it returned an average of about 243 context tokens per question. See the benchmarks for more details: https://github.com/bit-robotics/recalld-benchmarks

Recalld can be accessed via API or MCP. Benchmarks are public and can be rerun.

Recalld has a free tier plan with no card required. We offer two completely separate regions, EU and US, depending on where you want your data to be stored.

Recalld is hosted only and cannot be self-hosted yet