Knowledge Management

HiveMind OS builds a persistent knowledge graph from your conversations. You can teach it explicitly, let it learn automatically, and query it at any time.

Teaching the Agent

Tell the agent what you want it to remember using natural language in the chat:

Remember that I prefer TypeScript with strict mode
Our API uses REST with JSON, auth via JWT tokens
The production database is PostgreSQL 15 on AWS RDS

The agent uses its internal memory system to store facts as nodes in the knowledge graph. You don't need any special commands — just ask naturally and the agent will persist important context.

TIP

Be specific. "Remember that our deploy target is ECS Fargate in us-east-1" is more useful than "We use AWS".

Automatic Knowledge Extraction

During context compaction HiveMind OS automatically extracts entities, relationships, and facts from your conversations — no action required. Over time the graph grows to include technologies, architecture decisions, team conventions, and more.

Searching and Browsing Knowledge

Natural Language Queries

Ask the agent to recall information in natural language:

What database do we use in production?
What technologies have we discussed in our architecture sessions?

The agent uses the knowledge.query tool internally to search the knowledge graph. This tool supports two actions:

• search — full-text search across knowledge nodes
• explore — retrieve a specific node and its neighbors in the graph

Knowledge Explorer UI

You can also browse the full knowledge graph visually in the Knowledge Explorer — a dedicated UI view that lets you navigate nodes, relationships, and clusters interactively.

Knowledge Scopes and Namespaces

Knowledge is organized into scopes so the right facts surface in the right context:

Scope	Applies to	Example
Global	All workspaces and sessions	General preferences like UI theme
Workspace	A specific project directory	Project-specific tools and conventions

Workspace-scoped knowledge activates automatically when you open a session inside that project. Facts remembered while working in a project are automatically scoped to that workspace.

INFO

Shared namespaces let teams pool knowledge across workspaces. See Personas for namespace configuration.

Classification and Data Protection

Every knowledge node inherits the data classification of the conversation it came from:

Classification	Behavior
PUBLIC	Available to all providers
INTERNAL	Available to internal providers
CONFIDENTIAL	Prompted before sharing externally
RESTRICTED	Never sent to external providers

This means a fact learned in a RESTRICTED session will never leak to a PUBLIC provider — classification travels with the knowledge automatically.

WARNING

If a workspace is classified as RESTRICTED, all facts remembered inside it inherit that classification. Review workspace classification in Settings → Workspace → Classification, or right-click specific files/folders in the Workspace browser to set a manual classification override.

Agent Kits

To share agent configurations across environments, use Agent Kits. Kits export personas, workflows, skills, and attachments as a portable .agentkit ZIP file — but not knowledge graph data. Knowledge is managed separately per instance.

See the Agent Kits guide for full details on exporting, importing, and namespace remapping.

Next Steps

• Personas — create specialized agents that leverage stored knowledge
• MCP Servers — connect external tools the agent can remember how to use
• Privacy & Security — learn more about data classification