Embedded local-first graph memory for AI agents.

Quick Start · Default Stack · How It Works · MCP · HTTP API · Config

MollyGraph is a local graph-and-vector memory service for agents.

The default path is simple:

• ingest text
• extract entities and relationships with GLiNER2
• store graph and vectors locally in Ladybug
• query through MCP, HTTP, or the Python SDK

The goal is not “a vector store with a wrapper.” The goal is structured local memory that agents can actually use.

git clone https://github.com/brianmeyer/mollygraph.git
cd mollygraph
./scripts/install.sh
./scripts/start.sh

The canonical local runtime is Python 3.12. ./scripts/install.sh will use python3.12 when it is available, or uv to provision Python 3.12 automatically.

The install script:

• creates service/.env from service/.env.example if needed
• creates the runtime venv at service/.venv
• preloads the default GLiNER2 and embedding models when possible

Default local API:

• base URL: http://127.0.0.1:7422
• auth: Bearer dev-key-change-in-production

Production-style smoke test:

service/.venv/bin/python scripts/production_smoke.py --json

• graph storage: Ladybug
• vector storage: Ladybug
• extraction: GLiNER2
• default embeddings: Snowflake/snowflake-arctic-embed-s
• queue: SQLite
• client surfaces: MCP, HTTP, SDK

The default runtime does not require Neo4j. Neo4j-backed audit, training, and decision-oriented flows are optional compatibility surfaces, not the base product.

text in
  -> SQLite extraction queue
  -> GLiNER2 extraction
  -> speaker anchoring + relation gate
  -> Ladybug graph + Ladybug vector storage
  -> MCP / HTTP / SDK query surface

Default retrieval path:

• graph exact-match and vector similarity run in parallel
• results are merged into one response

The current graph and vector layers use separate .lbug files by default. That keeps the runtime predictable while the shared-Ladybug storage path is evaluated.

Example MCP config:

{
  "mcpServers": {
    "mollygraph": {
      "command": "mollygraph-mcp",
      "args": ["--base-url", "http://localhost:7422", "--api-key", "YOUR_KEY"]
    }
  }
}

Default tools:

• add_episode
• search_facts
• search_nodes
• get_entity_context
• get_queue_status
• delete_entity
• prune_entities

Optional audit or training tools are only exposed when the running backend reports support for them.

pip install "git+https://github.com/brianmeyer/mollygraph.git#subdirectory=sdk"

from mollygraph_sdk import MollyGraphClient

client = MollyGraphClient(
    base_url="http://localhost:7422",
    api_key="YOUR_KEY",
)

client.ingest("Sarah joined the ML team at Acme Corp.", source="slack")
result = client.query("What do we know about Sarah?")
entity = client.get_entity("Sarah")
client.close()

Core endpoints:

Operator utilities still exist for quality checks, vector reconciliation, and embedding refresh, but they are not the main onboarding path.

Everything is env-driven. See service/.env.example for the full configuration surface.

Key defaults:

MOLLYGRAPH_GRAPH_BACKEND=ladybug
MOLLYGRAPH_VECTOR_BACKEND=ladybug
MOLLYGRAPH_EXTRACTOR_BACKEND=gliner2
MOLLYGRAPH_EMBEDDING_ST_MODEL=Snowflake/snowflake-arctic-embed-s
MOLLYGRAPH_EMBEDDING_VECTOR_DIMENSION=384

Optional local alternate:

MOLLYGRAPH_EMBEDDING_OLLAMA_MODEL=nomic-embed-text

Current docs:

• docs/ARCHITECTURE.md
• docs/DOCS_MAP.md
• docs/PRODUCTION_TEST_CHECKLIST.md
• service/README.md
• sdk/README.md
• service/BACKLOG.md

Later-phase plan:

• service/DECISION_TRACES_PLAN.md

MIT