Use docs/DOCS_MAP.md first so you know which markdown files are current and which are experimental plans.

1. Install dependencies:

./scripts/install.sh

This creates service/.venv, pins it to Python 3.12, and bootstraps service/.env from the example file if needed. If python3.12 is not already installed, the script can use uv to provision it.

2. Configure the service if you want to change defaults:

$EDITOR service/.env

3. Start the default local-first stack:

./scripts/start.sh

Neo4j is optional and only needed if you are explicitly working on legacy or experimental surfaces that still depend on the old backend.

Use the project service venv:

service/.venv/bin/python -m pytest -q

That root command runs both tests/ and service/tests/.

Useful focused runs:

service/.venv/bin/python -m pytest service/tests/test_ladybug_core_flow.py -q
service/.venv/bin/python -m pytest service/tests/test_graph_quality.py -q
service/.venv/bin/python -m pytest service/tests/test_ladybug_graph.py service/tests/test_ladybug_vector_backend.py -q

Run startup script preflight checks:

bash -n scripts/install.sh scripts/start.sh

Validate SDK + MCP adapter packaging:

pip install -e "sdk[mcp]"
mollygraph-mcp --help
python -c "from mollygraph_sdk import MollyGraphClient; print('sdk ok')"

Protected endpoints require bearer auth. Default local key: dev-key-change-in-production.

• Keep the default runtime local-first.
• Keep the core product path small: ingest, query, entity context, cleanup, health, and stats.
• Treat audit, training, decision traces, and extra enrichment layers as optional unless the task is explicitly about them.
• Do not reintroduce cloud-only requirements into the default product path.