DOCKER + MCP STACK ONE-TIME $99 • SELF-HOSTED
Your AI agents forget everything. Give them a private memory.
AgentContext Vault is a Docker stack you install on your own VPS. Antigravity, Cursor, Claude Desktop, Hermes and n8n read and write the same project memory. Nothing is sent to a third-party memory vendor.
No subscription. Runs on your own 4 GB Linux VPS. 14-day refund by email.
→ Postgres with row-level security (decided Mon)
→ Auth: Argon2id + TOTP, done
→ Rule: conventional commits
invoices is still missing. I will add it first.load_context is a real MCP tool the stack provides.One memory, every tool you use
Most agent memory lives in a vendor's cloud or trapped inside one app's chat history. This keeps it in one place that you control and every agent can reach.
Agents load it, then save to it
At the start of a session an agent calls load_context. During work it calls save_conversation and save_decision. Fourteen MCP tools cover projects, rules, search and credentials.
You can inspect what it knows
The included React dashboard shows projects, conversations, decisions and rules, plus a live graph of how they connect, so you can verify what your agents recall.
It stays on your server
Postgres, Neo4j and Redis run in your containers. Credentials are encrypted with a Fernet key only you hold. Client specs never touch a SaaS memory provider.
See how your projects connect
The dashboard's graph view links each project to the decisions, rules, technologies and conversations behind it. Select any node below to inspect it.
Example data drawn to match the dashboard's node types. Your own graph builds automatically from what your agents save.
Install in an afternoon, not a weekend
You need a Linux VPS and, for the dashboard, a domain pointing at it.
Run the installer
Installs Docker if missing, generates every secret into .env, builds the stack, creates your admin user and prints your URLs.
unzip agentcontext-vault.zip cd agentcontext-vault sudo ./setup.sh
Point your tools at it
Copy the config for your tool, fill in your domain and API key. Ready-made configs for Antigravity, Cursor, Claude Desktop, Hermes and n8n are included.
{
"mcpServers": {
"vault": {
"serverUrl": "https://memory.yourdomain.com/mcp/sse",
"headers": { "x-api-key": "vault_..." }
}
}
}
Tell the agent when to save
One line in your agent's rules file. From then on it loads context at the start and saves as it goes.
# in your rules / system prompt Call load_context at the start of every session. Call save_conversation every 15 messages and save_decision after each technical decision.
What is in the zip
Full source, not a hosted account. Read it, change it, run it on your own hardware.
agentcontext-vault/ ├─ README.md install, connect, troubleshoot ├─ setup.sh one-command installer ├─ backup.sh Postgres + Neo4j + .env backups ├─ docker-compose.yml 6 services, 1 optional ├─ .env.example ├─ api/ FastAPI: REST, MCP over SSE, auth ├─ dashboard/ React: graph, projects, vault ├─ caddy/ automatic HTTPS ├─ connect/ Antigravity, Cursor, Claude, Hermes, n8n └─ docs/ API reference, architecture
- Services
- Neo4j, Postgres, Redis, the FastAPI backend, the React dashboard and Caddy. Ollama is an optional seventh for entity extraction.
- Dashboard login
- Argon2id passwords, HTTP-only session cookies and TOTP two-factor sign-in.
- Credential vault
- Fernet-encrypted entries your agents can store and read back over MCP or REST.
- Backups
- A cron-ready backup script with 14-day rotation.
- Licence
- Single organisation. Modify it for your own agency and your clients' projects. No resale or redistribution of the source.
Who this is for, and the limits
A good fit if you
- run AI agents for clients and cannot park their specs in a vendor's cloud
- use more than one agent tool and want them to share context
- are comfortable running
docker composeon a Linux VPS - would rather pay once than add another monthly SaaS bill
Know before you buy
- Automatic entity extraction uses a local Ollama model (~8 GB RAM). Without it, saving, search, context, rules, vault and all 14 MCP tools still work, but entities are not auto-extracted.
- The dashboard login cookie requires HTTPS, so use a domain pointing at your VPS. A bare server IP will not log in to the web UI.
- One-time download. Questions and bug reports go to [email protected].
SINGLE-ORG LICENSE
One payment. Your server, your data.
Hosted memory services bill every month for as long as your agents run. This is a single purchase you own and host yourself.
- Full source for the FastAPI server, React dashboard and installer
- Ready-made configs for Antigravity, Cursor, Claude Desktop, Hermes & n8n
- Caddy HTTPS config, backup script and full API docs
Pay with PayPal or a credit/debit card through PayPal checkout. Instant zip download unlocked right after capture. 14-day refund policy.
Questions
How much Docker or Linux do I need to know?
Enough to SSH into a server and run ./setup.sh. The installer handles Docker, secrets, databases and HTTPS. The README covers common troubleshooting.
Which AI tools can connect?
Anything that speaks MCP over SSE or can call a REST API. Ready-made configs are included for Antigravity, Cursor/Windsurf, Claude Desktop, Hermes Agent and n8n.
What VPS do I need?
Ubuntu 22.04+ or Debian 12+, 2 vCPU and 4 GB of RAM for the core stack. Plan for 8 GB if you enable local Ollama entity extraction.
Do my agents' conversations leave my server?
The stack sends nothing out. Your tools still talk to whichever LLM providers you configure for chat, while all saved memory and encrypted credentials stay strictly on your VPS.
What happens if I lose the encryption key?
Stored credentials cannot be decrypted without VAULT_KEY. The installer writes it to .env, and backup.sh copies that file.
Can I use it for my clients?
Yes. The licence covers one organisation, including work on your clients' projects. You may not resell or redistribute the source.
What if it does not work for me?
Email [email protected]. If you cannot get it running, request a full refund within 14 days by email.