# MemHouse: Governed Memory Engine for AI Agents & Swarms
> The stateful, auditable memory layer for personalized consumer AI, personal agents (OpenClaw, Hermes Agent), orchestrators (LangGraph, CrewAI, AutoGen), and ephemeral agent swarms.

## Overview
MemHouse is an open-source, highly concurrent memory governance engine backed by PostgreSQL/pgvector. Unlike raw vector databases that blindly index embeddings without governance, MemHouse subjects extracted facts to strict isolation, Gate A/Gate B quality filtering, and human-in-the-loop curation states (Held → Active).

## Core Application: Personalized Consumer AI
MemHouse solves the two fundamental problems of personalized AI at scale:

1. **Peer Isolation Violation** — A shared vector database cannot safely store per-user memory. Without strict account-level isolation, user A's conversation history and preferences leak into user B's context. MemHouse uses cryptographic account-scope boundaries to guarantee zero cross-user data leakage.

2. **Context Window Blowout** — You cannot inject 3 years of conversation history into every prompt. MemHouse extracts durable facts (preferences, behavioral patterns, history) and serves them on-demand without raw log injection.

**Dual-Scope Memory Model:**
- Each user's conversation history, preferences, and behavioral patterns live in a cryptographically isolated personal scope.
- Product knowledge, FAQs, policies, and shared catalog data live in a global scope accessible to all agents.
- Both scopes are merged per-query — peers never see each other's data.

Agents remember conversation history, long-term peer preferences, and behavioral patterns across sessions.

## Key Capabilities
- **Strict Peer Isolation**: Hard Account and Scope boundaries. User facts never leak across accounts or peers.
- **5-Step Stateful Pipeline**: Raw Observation → Extraction → Gate A Policy Check → Gate B Blast Radius Check → Governed Knowledge Store.
- **Dual-Scope Memory**: Combine isolated per-user scopes with a shared internal knowledge base in a single query.
- **Ecosystem Fit**: Native connectors for OpenClaw CLI, Hermes Agent, LangGraph Checkpointers, CrewAI Shared Memory, AutoGen, LlamaIndex, and Claude Desktop MCP.
- **Process Death-Proof**: External persistent state memory for short-lived ephemeral workers and agent swarms.
- **100% Auditable**: Tracks full provenance metadata, confidence scores, and source turn IDs for every fact.

## Use Cases
- **Personalized Consumer AI** — Per-user memory with strict peer isolation + shared product knowledge base. See: https://memhouse.org/use-cases/personalized/
- **Customer Support Copilots** — Tenant-scoped ticket history and procedural golden responses. See: https://memhouse.org/use-cases/support/
- **Regulated Financial & Legal AI** — Human-in-the-loop curation and exact provenance citations. See: https://memhouse.org/use-cases/regulated/
- **Personal Agents (OpenClaw, Hermes)** — Persistent local memory across CLI sessions without credential leakage.
- **Ephemeral Agent Swarms** — External state store for LangGraph, CrewAI, AutoGen swarms that die between tasks.

## API & MCP Specifications

### Model Context Protocol (MCP) Tools
MemHouse exposes standard MCP tools for direct integration with Claude Desktop, Cursor, OpenClaw, and custom LLM agents:

1. `mcp_memhouse_save_fact`
   - Description: Save an extracted fact into governed memory.
   - Parameters:
     - `account_id` (string, required): Unique account identifier.
     - `scope` (string, required): Domain scope (e.g., `/user/preferences`, `/codebase/architecture`).
     - `fact_text` (string, required): Natural language statement of the fact.
     - `confidence` (number, optional): Initial confidence score between 0.0 and 1.0.

2. `mcp_memhouse_query_context`
   - Description: Retrieve salient governed facts relevant to the user query.
   - Parameters:
     - `account_id` (string, required): Unique account identifier.
     - `query` (string, required): Conversational context or question.
     - `limit` (number, default: 5): Maximum facts to retrieve.

3. `mcp_memhouse_approve_fact`
   - Description: Approve a Held state fact into Active memory.
   - Parameters:
     - `fact_id` (string, required): Fact UUID.

### REST API Endpoints
- `POST /api/v1/observations`: Ingest raw conversational turn payload.
- `GET /api/v1/context`: Query salient active facts.
- `GET /api/v1/facts?state=held`: List facts awaiting human curation.
- `POST /api/v1/facts/:id/approve`: Promote fact state to active.

## Blog & Release Notes
All MemHouse technical articles, benchmarks, and release notes are published at https://memhouse.org/blog/ and maintained as MDX files in `src/content/blog/`:

- `src/content/blog/release-v0-4-0-ephemeral-swarms-and-mcp.mdx` — Release notes for v0.4.0: Ephemeral Swarms & MCP tool server. URL: https://memhouse.org/blog/release-v0-4-0-ephemeral-swarms-and-mcp/
- `src/content/blog/achieving-10k-ingests-per-second-pgvector.mdx` — Benchmark: 10,000 ingests/sec with pgvector. URL: https://memhouse.org/blog/achieving-10k-ingests-per-second-pgvector/
- `src/content/blog/memhouse-vs-mem0-architectural-differences.mdx` — Architectural comparison: MemHouse vs Mem0. URL: https://memhouse.org/blog/memhouse-vs-mem0-architectural-differences/
- `src/content/blog/building-support-copilot-in-10-minutes.mdx` — Tutorial: Build a governed support copilot. URL: https://memhouse.org/blog/building-support-copilot-in-10-minutes/

## Latest Release: v0.4.0 (2026-08)
- **Features**: Ephemeral Swarm Persistence, OpenClaw MCP Tool Server, Gate B Threshold Curation UI, High Concurrency Pool scaling to 10,000 ingests/sec.

## Site Pages
- Home: https://memhouse.org/
- Solutions (Security): https://memhouse.org/solutions/security/
- Solutions (Engineers): https://memhouse.org/solutions/engineers/
- Use Cases (Personalized): https://memhouse.org/use-cases/personalized/
- Use Cases (Support): https://memhouse.org/use-cases/support/
- Use Cases (Regulated): https://memhouse.org/use-cases/regulated/
- Blog: https://memhouse.org/blog/
- Resources: https://memhouse.org/resources/
- Startup Program: https://memhouse.org/startup-program/
- Contact: https://memhouse.org/contact/
- Privacy Policy: https://memhouse.org/privacy/
- Terms of Service: https://memhouse.org/terms/
- Compliance Whitepaper: https://memhouse.org/compliance/

## Documentation & Links
- Website: https://memhouse.org
- Docs: https://memhousehq.github.io/memhouse/
- GitHub: https://github.com/memhousehq/memhouse
- Helm Chart: `helm repo add memhouse https://charts.memhouse.org`
- Docker Image: `ghcr.io/memhousehq/memhouse:latest`