Graphiti agent memory on embedded Ladybug, one file per user

[renezander030]

I maintain graphiti-local, an independent project (not affiliated with LadybugDB or Zep) that
runs Graphiti's temporal knowledge graph on embedded Ladybug instead of Neo4j or FalkorDB.

What it does with Ladybug:

• One .ladybug file per group/user, so a person's memory can be backed up, moved or deleted
  as a single file (related to the multi-file question in Ladybug multi-file storage feature #1145).
• Facts keep validity times; a newer fact retires the old one instead of overwriting it.
• Agents read over MCP (six read-only tools). Proposed changes wait in a review queue and land
  only after a human approves them.
• Ollama for extraction and embeddings, so the whole stack is local.

Try the read path in about 10 seconds, no model needed:

uvx --from git+https://github.com/renezander030/graphiti-local kg-demo

Repo: https://github.com/renezander030/graphiti-local
Measured results and known limits (7B extraction does make mistakes):
https://github.com/renezander030/graphiti-local/blob/main/docs/local-demo-results.md

Feedback on how I use the Python driver is welcome, especially around schema setup on an
empty graph (I hit a Binder exception on a graph with no edges, since fixed on my side).