Hypabase

A Python hypergraph library with provenance and SQLite persistence.

Install

uv add hypabase

Related MCP server: memory-mcp

Quick example

from hypabase import Hypabase

hb = Hypabase("my.db")

# One edge connecting five entities
hb.edge(
    ["dr_smith", "patient_123", "aspirin", "headache", "mercy_hospital"],
    type="treatment",
    source="clinical_records",
    confidence=0.95,
)

# Query edges involving a node
hb.edges(containing=["patient_123"])

# Find paths between entities
hb.paths("dr_smith", "mercy_hospital")

Features

• Hyperedges — an edge connects 2+ nodes in a single relationship

• Provenance — every edge carries source and confidence

• SQLite persistence — data persists to a local file automatically

• O(1) vertex-set lookup — find edges by their exact node set

• Namespace isolation — .database("name") for scoped views in a single file

• Provenance queries — filter by source and min_confidence, summarize with sources()

• Memory MCP server — 7 tools for AI agent persistent memory (remember, recall, forget, consolidate, connections, who_knows_what, resolve_contradiction)

• CLI — hypabase init, hypabase node, hypabase edge, hypabase query

Provenance

Every edge carries source and confidence:

hb.edge(
    ["patient_123", "aspirin", "ibuprofen"],
    type="drug_interaction",
    source="clinical_decision_support_v3",
    confidence=0.92,
)

# Bulk provenance via context manager
with hb.context(source="schema_analysis", confidence=0.9):
    hb.edge(["a", "b"], type="fk")
    hb.edge(["b", "c"], type="fk")

# Query by provenance
hb.edges(source="clinical_decision_support_v3")
hb.edges(min_confidence=0.9)

# Overview of all sources
hb.sources()

Namespace isolation

Isolate data into separate namespaces within a single file:

hb = Hypabase("knowledge.db")

drugs = hb.database("drugs")
sessions = hb.database("sessions")

drugs.node("aspirin", type="drug")
sessions.node("s1", type="session")

drugs.nodes()     # -> [aspirin]
sessions.nodes()  # -> [s1]

What is a hypergraph?

In a regular graph, an edge connects exactly two nodes. In a hypergraph, a single edge — called a hyperedge — can connect any number of nodes at once.

Consider a medical event: Dr. Smith prescribes aspirin to Patient 123 for a headache at Mercy Hospital. In a traditional graph, you'd split this into binary edges — doctor-patient, doctor-drug, patient-hospital — and the fact that they belong to one event becomes an inference, not a structure. A hypergraph stores this natively: one edge connecting all five entities.

This matters because real-world relationships often involve more than two things. A paper has three or four authors, not one. A transaction involves a buyer, a seller, a product, and a payment method. A chemical reaction has reagents and products on both sides. Forcing these into pairs means the grouping becomes implicit.

Why provenance?

When relationships come from different sources — manual entry, LLM extraction, sensor data, clinical records — you need to know where each one came from and how much you trust it. Hypabase tracks this with two fields on every edge: source (a string identifying the origin) and confidence (a float from 0 to 1). You can filter queries by these fields and get a summary of all sources in your graph with hb.sources().

Where hypergraphs show up

• Knowledge graphs — representing complex real-world relationships without decomposition

• Agent memory — structured, queryable memory for AI agents that persists across sessions

• Biomedical data — drug interactions, clinical events, molecular pathways

• RAG pipelines — storing extracted relationships for retrieval-augmented generation

• Supply chains, collaboration networks, and anywhere relationships involve more than two things

The broader idea has roots in AI research going back to OpenCog's AtomSpace, which uses hypergraph-like structures to represent knowledge for AGI. More recent work applies hypergraphs specifically to retrieval and reasoning:

• HyperGraphRAG — n-ary knowledge retrieval across medicine, agriculture, CS, and law

• Cog-RAG — dual-hypergraph retrieval with theme-level and entity-level recall

• Hypergraph Memory for Multi-step RAG — hypergraph-based memory for long-context relational modeling

MCP server

Hypabase includes a Memory MCP server with 7 tools so AI agents can use it as structured, persistent memory. Works with Claude Code, Claude Desktop, Cursor, Windsurf, and any MCP-compatible client.

uv add hypabase
hypabase-memory

CLI

uv add hypabase
hypabase init
hypabase node dr_smith --type doctor
hypabase edge dr_smith patient_123 aspirin --type treatment --source clinical_records
hypabase query --containing dr_smith
hypabase stats

Documentation

docs.hypabase.app

License

Apache 2.0