Overview

What Hive is, and the one architectural decision that matters most.


Aiqbee Hive is a self-hosted knowledge graph designed to be read and written by AI agents. It runs as a single container with PostgreSQL behind it, and exposes itself two ways:

  • MCP (Model Context Protocol) — 34 tools an LLM can call directly. This is the primary interface, and the one to reach for when building an agent or assistant.
  • REST (/api/v1) — a small management API for provisioning and automation: create brains, list them, patch settings. Not intended for agent traffic.

The vocabulary

TermWhat it is
Brain A knowledge graph. Access is granted per brain, and an MCP connection is scoped to exactly one.
Neuron One unit of knowledge — a title plus content, optionally dated. Embedded on write so it is semantically searchable.
Neuron type A category (Decision, Meeting Note, Standard). Types can carry their own access rules under Advanced Security.
Synapse A described link between two neurons. The description matters — it is what lets an agent explain why two things relate.
Service account A non-human identity with its own client credentials. What your integration should authenticate as.

How a brain gets its shape

Neuron types are the schema, and they are per-brain. Before writing, an agent should call aiqbee_get_neuron_types and use a real type id — they are not guessable, and a wrong id is rejected. Types can be created on the fly with aiqbee_add_neuron_type, but a brain with forty near-duplicate types is worse than one with eight good ones, so prefer reusing what is there.

Where to go next