Medulla

The medulla distribution provides Medulla; import it as epistemic_graph. The repository is medulla. Package, CLI and API identifiers remain unchanged.

A durable graph database and reasoning engine for connected, evidence-rich data.
_{Property graph · SQL · RDF/OWL · vectors · time · provenance · multimodal records}

Documentation · Capabilities · Interfaces · Status

Overview

Medulla is a Rust-native database and compute engine for teams building knowledge systems, evidence-led applications, analytics, and agent memory. It stores and queries connected graph records, relational data, RDF/OWL knowledge, vectors, time-series, events, documents, and media through one engine.

Claims can retain their evidence, provenance, confidence, and time of validity, so applications can inspect why information is present and how it changes. Use Medulla directly through its clients and query interfaces. It also serves as the durable knowledge layer beneath Arbora and Agent Utilities. Medulla is the database. Agent execution and connector runtimes live in their respective ecosystem projects.

Key capabilities

Check the capability matrix and generated method ledger for the current behavior of each operation.

Packages

This repository publishes two PyPI packages and one container image, all at the same version.

Medulla Stem is the native server, distributed as medulla-stem. The server executable, module and packaging-path identifiers remain unchanged.

Documentation

• Start here for a guided engine tour.
• Interfaces explains UQL, SQL, SPARQL, Cypher, GraphQL, vectors, time-series, and clients.
• Architecture covers query execution, reasoning, storage, and distribution.
• Deploy Medulla covers durable server, TLS, containers, and clusters.
• Operations covers day-two procedures and recovery.

Architecture

Explore the live architecture

People use the Agent Web UI, Agent Terminal UI, Geniusbot, or the messaging services hosted by Arbora. The terminal UI reaches Arbora through its REST API. External applications and agents connect through Arbora using MCP, REST, or A2A. Arbora governs the runtime boundary. Agent Utilities coordinates agents and workflows. Medulla stores and reasons over their durable knowledge. Source systems enter through Agent Connector SDK, which commits typed source data to the graph.

Quick start

Install the release wheel and open an in-process graph. This mode is explicit, ephemeral, and needs no server or TLS configuration.

python -m pip install medulla
python -m epistemic_graph.cli status  # no server is expected in embedded mode
python - <<'PY'
import msgpack
from epistemic_graph.engine import Engine

engine = Engine(persist_dir=":memory:")
engine.create_graph("demo")
engine.add_node(
    "demo",
    "node:hello",
    msgpack.packb({"kind": "Greeting"}, use_bin_type=True),
)
print(engine.has_node("demo", "node:hello"), engine.node_count("demo"))
PY

Expected output:

Status: NOT RUNNING (no PID file)
True 1

For durable storage, authenticated clients, containers, TLS, or clustering, continue with Deploy Medulla.

Contributing

Issues and pull requests are welcome. See CONTRIBUTING.md for development setup and review gates. Architecture and implementation guidance for coding agents lives in AGENTS.md.

License

Licensed under the MIT License.