no_llm

Run Without an LLM

Air-gapped or budget-conscious? Use neo4j-agent-memory with no LLM provider — local embeddings, local NER, no API keys.

This example shows how to wire MemorySettings for environments where you can't (or don't want to) call an LLM. The key is llm=None plus a non-LLM extractor and a local embedder.

⚠️ Neo4j Labs Project

This example is part of neo4j-agent-memory, a Neo4j Labs project. It is actively maintained but not officially supported. APIs may change. Community support is available via the Neo4j Community Forum.

When to use this

• Air-gapped or offline deployments where outbound API calls aren't allowed.
• Cost-sensitive workloads where every LLM call counts.
• Deterministic test environments where you want zero variability from a remote model.
• Bootstrapping a new project before you've decided on an LLM vendor.

What this demonstrates

• llm=None — explicit opt-out. Validated at construction time.
• Provider-string shorthand for local embeddings — "sentence-transformers/all-MiniLM-L6-v2" resolves to SentenceTransformersProvider via from_provider. No external API calls.
• ExtractorType.PIPELINE with enable_llm_fallback=False — multi-stage spaCy + GLiNER pipeline, no LLM rescue.
• Configuration-time validation — if you pair llm=None with an extractor that requires an LLM, MemorySettings raises a ValidationError naming both fields rather than failing later at runtime.

settings = MemorySettings(
    neo4j=Neo4jConfig(...),
    llm=None,                                          # explicit opt-out
    embedding="sentence-transformers/all-MiniLM-L6-v2",# local embeddings
    extraction=ExtractionConfig(
        extractor_type=ExtractorType.PIPELINE,
        enable_spacy=True,
        enable_gliner=True,
        enable_llm_fallback=False,            # required when llm=None
    ),
)

Prerequisites

pip install "neo4j-agent-memory[extraction,sentence-transformers]"
python -m spacy download en_core_web_sm

A running Neo4j 5.x. Set NEO4J_URI / NEO4J_PASSWORD if you're not using bolt://localhost:7687 with the default password.

Run

python main.py

You'll see a printed memory context built without a single call to OpenAI, no API key required.

Going further

• How-to guide: docs/modules/ROOT/pages/how-to/running-without-an-llm.adoc.
• Reference: docs/modules/ROOT/pages/reference/extractors.adoc — extractor menu, when to add LLM rescue, GLiNER schema choices.

Support

• 💬 Neo4j Community Forum
• 🐛 GitHub Issues
• 📖 neo4j-agent-memory documentation

---

Verified against neo4j-agent-memory v0.1.2 / v0.2-dev on 2026-05-03.