Anyone with write access to your store can insert embeddings indistinguishable from yours. Mnemo attaches identity at ingest and verifies it at retrieval, so your retriever ignores anything it did not produce.
If you embed documents and retrieve them at query time, Mnemo applies. Works with your existing system, including Pinecone, Weaviate, Qdrant, Milvus, and pgvector.
02Why it matters
Anyone can write embeddings. Verification is what matters.
If a document is in the index, the retriever returns it. There is no way to confirm it came from your pipeline. Enterprise buyers now ask for proof that one customer's vectors cannot be retrieved by another, and configuration alone does not answer that question.
03What this does
Two calls. Works with your system.
On insert, embed() attaches a signed identity. On retrieval, verify() returns whether the embedding came from your pipeline and whether it has been altered. Your model and your storage stay where they are.
04Scope
What Mnemo does not do.
Does not detect prompt injection
Does not analyze model weights
Does not moderate content
Does not replace your system
Mnemo focuses on one layer: attaching identity to embeddings and verifying them when they are used.
05Who this is for
VP ML Platform. Head of AI Infrastructure. Head of Security for customer-facing RAG.
06Why teams adopt it
Retrieval poisoning is a real risk in production pipelines.
Enterprise contracts ask for verifiable customer isolation.
Per-source provenance is required in regulated verticals.
Programmatic verification removes coordination cost between teams.
Start verifying in minutes.
Free tier covers small pipelines. The first verify is on us.