Skip to content

SKILL.md

Latest commit

 

History

History
53 lines (38 loc) · 1.88 KB

SKILL.md

File metadata and controls

53 lines (38 loc) · 1.88 KB

MollyGraph Skill

Name: mollygraph
Description: Local-first graph memory for AI agents
Version: 3.0.0

Overview

Use these docs as the source of truth:

MollyGraph's default product path is:

  • Ladybug for local graph storage
  • Ladybug for local vector storage
  • GLiNER2 for entity and relation extraction
  • Snowflake/snowflake-arctic-embed-s for default local embeddings
  • SQLite for the async extraction queue
  • HTTP + MCP + Python SDK for the user-facing surface

Neo4j, audit flows, training loops, and decision traces still exist as optional compatibility or later-phase surfaces, but they are not part of the default runtime.

Installation

cd /Users/brianmeyer/mollygraph
./scripts/install.sh
./scripts/start.sh

./scripts/install.sh creates service/.env if needed and installs the runtime into service/.venv. This starts the HTTP API on port 7422. The default API key is dev-key-change-in-production.

Default Runtime

MOLLYGRAPH_GRAPH_BACKEND=ladybug
MOLLYGRAPH_VECTOR_BACKEND=ladybug
MOLLYGRAPH_EMBEDDING_ST_MODEL=Snowflake/snowflake-arctic-embed-s

Graph and vector use separate Ladybug database files by default.

Working Guidance

  • Prefer the default surfaces: ingest, query, get_entity, queue status, health, and stats.
  • Treat audit, training, decision, and registry surfaces as explicitly optional.
  • Keep docs aligned with the default runtime before documenting later-phase features.
  • Use the service health capabilities to decide whether optional MCP tools should be exposed.
  • Default local data lives under ~/.graph-memory.