MemHQ is hosted memory infrastructure for AI agents. Add, search, and synthesize memory with three endpoints.

Welcome to MemHQ

MemHQ is hosted memory infrastructure for AI agents. It turns unstructured conversation, documents, and events into a structured, queryable knowledge graph — extracting facts, preferences, decisions, and relationships, resolving contradictions over time, and surfacing the right slice of memory back to your agent on demand.

The surface area you'll spend most of your time in is small:

POST /v1/memhq/add — ingest a message, event, or document.
POST /v1/memhq/search — retrieve memories relevant to a query.
POST /v1/memhq/ask — get a synthesized answer with citations.

Everything else (projects, users, groups, ontology, webhooks) wraps those three.

Start here

Quickstart

Add your first memory and ask a question in 60 seconds.

Memory API

The three endpoints that power most agent integrations.

API Reference

Full OpenAPI 3.1 spec, rendered interactively.

Concepts

The mental model — graphs, episodes, memories, reconciliation.

Cookbook

Copy-pasteable recipes for chatbots, RAG, and webhook ingestion.

Benchmarks

LoCoMo scores and what the numbers mean.

How memory is shaped

A MemHQ project holds two complementary stores:

Per-user graphs — private memory scoped to a single end-user ID. The default home for anything personal: preferences, history, ongoing threads.
Group graphs — shared knowledge accessible to a defined set of users. Useful for team workspaces, customer accounts, or any context where multiple people read from the same memory.

Search and ask queries can target one, the other, or both. Storage-layer RBAC enforces who can read what — see RBAC.

What's next

If you're evaluating MemHQ, start with the Quickstart and the Benchmarks page. If you already have an API key, Memory API is the fastest path to a working integration.

Welcome