USE CASES

What you build when memory has structure.

Memuron is not a vector cache. It is a typed, linked, multi-tenant graph that decides what to keep and remembers across sessions. Five places that changes what your agents can do.

Start buildingBrowse use cases ↓
memory graph - live

Pick what you are building.

← → to switch
01 - AGENTS AND ASSISTANTS

Assistants that remember the user, not just the session.

When a user says "actually, make it blue," a naive store appends a second fact and your agent believes both. The Guardian retrieves the candidate, recognizes the conflict, and updates in place.

  • Create-vs-update decided by an LLM, not a similarity threshold
  • Memories persist across sessions, scoped per end-user
  • Recall by meaning with hybrid semantic + keyword search
POST /memuron/memories
guardian - write plan
wasUser prefers the accent color red.
UPDATE - conflict resolved
nowUser prefers the accent color blue.
UNDER THE HOOD

One query language for the whole graph.

Pipe semantic search through graph traversal and regex filters. Every call returns the result plus an execution trace.

memuron - graph-filesystem.v1
$ cwd /spaces/work
> ls | grep "auth" | semantic "rate limiting" | links
trace:
  ls - 312 nodes
  grep - 47 match /auth/
  semantic - top-8 by meaning
  links - +14 neighbors
OK 8 items - 22 edges - 41ms
WHY NOT JUST A VECTOR DB

The same input, two outcomes.

vector database
  • Stores "red" and "blue" as two facts and returns both
  • Flat top-k list, no relationships between results
  • You write chunking, dedup, and isolation yourself
memuron
  • Guardian merges the conflict into one coherent fact
  • Traversable graph links facts to sources and neighbors
  • Chunking, dedup, and tenant isolation built in

Build agents that actually remember.

Deploy as a monolithic API or wire it straight into your IDE over MCP. Free tier to start.

Get StartedSee pricing