All posts
·6 min read

Self-Hosted AI Agent Memory: Full Data Sovereignty

Enterprise teams need AI agent memory without sending data to third-party clouds. Agent-Memo's self-hosted deployment keeps everything on-premise — one Docker container, full data sovereignty.

Self-HostedEnterpriseData PrivacyOn-Premise

TL;DR: Agent-Memo now offers a self-hosted deployment for teams that need AI agent memory without sending data to external clouds. One Docker container, full admin dashboard, license-based activation. Your data never leaves your network.

Why self-hosted AI agent memory matters

AI coding agents like Claude Code, Cursor, and Windsurf are becoming essential developer tools. But they have a fundamental gap: no persistent memory. Every session starts from scratch.

Cloud memory services solve this problem, but many organizations can't use them. Regulated industries (finance, healthcare, defense), companies with strict data residency requirements, and teams working on proprietary code all need their AI agent memory to stay on their own infrastructure.

What makes self-hosted different

Agent-Memo's self-hosted deployment gives you the exact same 14 MCP memory tools as the cloud version, but running entirely on your servers:

  • Full data sovereignty — All memories, knowledge graph data, and embeddings stay in your PostgreSQL database
  • No external API calls — The BGE-M3 embedding model runs locally inside the container. No data sent to OpenAI, HuggingFace, or any third party
  • Air-gapped capable — Works in networks with no internet access after initial deployment
  • White-label ready — Customize the branding with your company name via the admin dashboard
  • Single container — Backend, frontend, and embedding model all packaged in one Docker image (~1GB)

How it works

The self-hosted deployment is a single docker compose up away:

  1. Run the install script on your server (Ubuntu, 4+ cores, 8GB+ RAM)
  2. Enter your license key in the setup wizard
  3. Create your admin account
  4. Configure company name and settings
  5. Start connecting agents via MCP

The setup wizard handles everything: database initialization, embedding model download, admin account creation, and initial configuration.

Admin dashboard

The self-hosted version includes a full admin dashboard for managing your deployment:

  • User management — Create, edit, and deactivate user accounts
  • Team management — Create teams, add members, assign roles
  • License management — View license status, expiry, and seat usage
  • System configuration — Company name, registration settings, SMTP

Same MCP tools, same agent experience

Developers using the self-hosted version have the identical experience to cloud users. The same .mcp.jsonconfiguration works — just point theAGENTMEMO_API_URL to your server instead of api.agent-memo.ai.

All 14 MCP tools are available:

  • memory_store, memory_recall, memory_overview — Core memory operations
  • kg_add, kg_query, kg_timeline — Knowledge graph
  • Claude Code hooks for auto-save (SessionStart, Stop, PreCompact)

Pricing

Self-hosted licenses are annual subscriptions, tiered by team size:

  • Starter (up to 25 users) — $3,000/year
  • Business (up to 100 users) — $8,000/year
  • Enterprise (unlimited users) — $20,000/year

All tiers include unlimited projects, memories, API calls, and knowledge graph facts. The license is verified offline — no phone-home, no usage tracking.

Contact sales for a license key, or view all plans to compare cloud vs self-hosted.

Related articles

Give your AI agent persistent memory today.

Get started free