Documents, queries, products, indicators. Whatever your pipeline already accepts.
Embed
Your model produces a vector. Mnemo attaches identity to it.
Store (optional)
Send the vector to Pinecone, Weaviate, Qdrant, pgvector, Milvus, or skip storage entirely.
Retrieve
Your retriever queries the store, or your service hands the vector back directly.
Verify
Mnemo confirms the result came from your pipeline and has not been altered.
Decision
Your service uses verified results. Unverified results are excluded.
02SDK
Two calls in your application.
Wrap your embedding step with embed(). Wrap your retrieval step with verify(). Your model and your storage stay where they are.
embed()
Attaches identity to a freshly produced vector.
WHEN
Once per ingest, after your model returns.
verify()
Returns identity and integrity for a vector you read back.
WHEN
On retrieval, before the result reaches the model or the user.
from mnemo import Mnemo
mn = Mnemo()
# on insert
vec = model.embed(doc)
e = mn.embed(vec, source="docs/policy.pdf")
store.upsert(e.vector)
# on retrieve
for r in store.query(q, top_k=5):
result = mn.verify(r.values)
if result.verified:
ctx.append(r)
03Trust Guardian
Verification for agents and services.
Trust Guardian lets agents and services verify embeddings through a shared MCP interface.