Frequently asked questions
What we get asked most about privacy, integration and how MemoryFirst fits your stack. Don't see your question? Get in touch.
The basics
What is MemoryFirst?+
The memory layer for AI: it ingests your business knowledge (emails, documents, calls, code, decisions), structures it into a fact graph and serves it to any agent via MCP or an OpenAI-compatible API, with verifiable citations.
How is it different from a vector database or a home-grown RAG?+
A vector store gives you similarity; MemoryFirst gives you a retrieved answer (dense + BM25 + graph, RRF fusion and reranking) with its source, contradiction resolution toward the most recent truth, and ingestion via connectors. It's not one piece — it's the whole memory layer.
Which models does it work with?+
It's model-agnostic. It exposes an OpenAI-compatible API and an MCP server (stdio, HTTP, SSE), so Claude, Cursor, ChatGPT or your own agent read it with no proprietary SDK. Swap models and your memory stays.
Privacy & data
Where does my data live?+
On European infrastructure (Hetzner). And it's self-hostable: you can run it in your own cloud if your compliance requires it.
Do you train models on my data?+
Never. Your information feeds neither our models nor third parties'. Your knowledge is yours.
Is it GDPR-compliant?+
Yes: GDPR-compliant processing, a data processing agreement (DPA), EU data residency and real retraction (a deleted source stops showing up in answers). Details on the security page.
How it works
How do citations work?+
Every answer includes the source: the paragraph in the document, the minute in the recording, the exact message. You debug retrieval instead of guessing, and your user can verify.
How does it stay current with contradicting information?+
Synthesis resolves toward the most recent truth instead of returning both versions, and retracted sources are excluded from retrieval.
What happens if the synthesis LLM is unavailable?+
The query degrades to the raw retrieved passages instead of failing. Your agent still gets useful context even when the generative layer has a bad day.
Integration & deployment
How do I integrate it?+
Via MCP (your assistant gains memory like any other tool) or the OpenAI-compatible API (swapping the base URL is usually most of the work). Start with the docs.
Can I self-host it?+
Yes. The components — API, vector database, embedder — run wherever you put them, with no phone-home. Same product, on your turf.
How does it compare to Mem0, Supermemory or Zep?+
Compare on what's hard to reverse: where the data lives, whether you can audit the answer and whether you can self-host. That's where we focus (EU, citations, MCP, no lock-in). If you want a fully managed SaaS with zero infrastructure, a cloud alternative will give you a faster hello-world.