MindGraph Documentation
MindGraph is a shared, governed memory layer for people and AI agents. It turns documents, conversations, connected systems, and agent work into durable knowledge with provenance, history, and scoped access.
Use the dashboard for day-to-day knowledge work and administration, or integrate through the HTTP API, TypeScript and Python SDKs, and MCP.
Quickstart
Create a workspace, add your first source, and make an API call.
Agent Memory Patterns
How to read and write memory: journals, ingestion, retrieval, sessions, and the complete agent loop.
Governed Agents
Configure long-running agents with scoped context, tool grants, policies, sessions, and schedules.
Build a Deal-Flow Agent
Power an investment pipeline with typed leads, source-backed diligence, safe updates, and deterministic scorecards.
Projects & Living Briefs
Scope a source corpus, ask grounded questions, and maintain a cited brief that changes with the evidence.
Governance & Access
Control what each user or agent can read, write, and do, with approvals and an audit trail.
MCP Connections
Connect external tools, grant them to agents explicitly, and govern consequential actions.
Knowledge Health
Inspect duplicates, contradictions, stale derivations, and incomplete analysis without silent mutation.
Operational Ontology
Define your business objects once, then fuse structured records with what documents and conversations say about them.
Authentication
API keys, dashboard sessions, organization scoping, and server-resolved principal identity.
SDKs
TypeScript and Python clients for the MindGraph Cloud API.
API Reference
The REST surface for graph operations, retrieval, projects, agents, governance, and integrations.
Concepts
Bitemporality, cognitive layers, typed relationships, retrieval, traversal, and multi-agent access.
Architecture at a glance
MindGraph runs on MnesticDB, its own bitemporal graph database engine forked from CozoDB. MindGraph records valid time on claims and relations, keeps an append-only version history for every node, and preserves source provenance; the engine's transaction-time axis and point-in-time queries are not yet exposed in the product. Above it, MindGraph adds ingestion, retrieval, identity-aware access, projects, governance, agent execution, and an optional operational ontology.