AI Toolkit and integrations
The Memgraph AI Toolkit is a set of open-source Python packages for building AI applications on Memgraph: knowledge graph creation, AI memory, GraphRAG and framework integrations. Each package installs on its own. Memgraph also works with AI frameworks maintained by other projects.
The diagram lists the packages in the toolkit and what each is for. The table below links each one to its docs.
AI Toolkit packages
Each package has its own page in these docs, except hygm and memgraph-toolbox, which are documented in their READMEs.
| Package | What it does | Docs |
|---|---|---|
| Context Graph | Gives coding agents memory across sessions and tools, stored in Memgraph. | AI memory |
| unstructured2graph | Extracts entities from documents into a knowledge graph, and relationships with an LLM or your own ontology. | Unstructured2Graph |
| structured2graph | Migrates a MySQL or PostgreSQL database into a graph. | Structured2Graph |
| hygm | Defines and validates a graph’s node and relationship types (its ontology). | README |
| mcp-memgraph | Connects MCP clients and agents to a Memgraph database. | Memgraph MCP server |
| langchain-memgraph | Queries Memgraph in natural language and builds LangChain agents. | LangChain |
| lightrag-memgraph | Uses Memgraph as LightRAG’s graph storage. | LightRAG |
| memgraph-toolbox | Runs Cypher and inspects a Memgraph database from Python. | README |
To report a bug or request a feature in any of these packages, open an issue on GitHub.
Third-party integrations
These frameworks maintain their own Memgraph integration:
- LlamaIndex: Create a knowledge graph from unstructured data and query it in natural language.
- Cognee: Extract concepts and relationships from text and search them as a knowledge graph.
- Mem0: Store and query AI agent memory as a graph, with Mem0 1.x.