Provides graph/relationship memory for GZMO, enabling memory to be stored and queried as a graph.
Provides a hot session/cache plane for GZMO memory, supporting fast session state and caching.
Provides durable local vault storage with full-text search for curated memory.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@gzmo-pi check my memory status"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
GZMO curated memory for Pi — Pi extension + MCP (via bundled pi-mcp-adapter).
Local-first by default (~/.gzmo SQLite vault). Optional living stack (Redis, Neo4j, Qdrant) documented below for operator hosts.
Gallery / install:
pi install npm:gzmo-piThen in Pi:
/gzmo setup
/reloadCall gzmo_memory_status / gzmo_memory_search.
Piece | Role |
This package | Pi extension ( |
pi-mcp-adapter (bundled) | Lazy MCP tools so context stays small |
GZMO binary | Product MCP server ( |
Product tools: gzmo_memory_status, gzmo_memory_search, gzmo_memory_turn_start, gzmo_memory_recall_pull, gzmo_memory_chain, gzmo_memory_profile, gzmo_wiki_search.
Related MCP server: engram
Install GZMO binary + init vault:
curl -fsSL https://raw.githubusercontent.com/maximilianwruhs-cyber/GZMO/main/scripts/install-gzmo.sh | bashInstall this Pi package:
pi install npm:gzmo-piWire MCP + verify:
/gzmo setup
/gzmo statusRestart Pi or /reload, then use gzmo_memory_status.
Manual MCP shape (if you prefer editing files):
{
"mcpServers": {
"gzmo-memory": {
"command": "/home/YOU/.local/bin/gzmo",
"args": ["mcp-serve"],
"env": {
"GZMO_CONFIG": "/home/YOU/.gzmo/gzmo.toml",
"GZMO_ALLOW_LAB_VAULT": "1",
"GZMO_PRODUCT": "1"
}
}
}
}Written to ~/.config/mcp/mcp.json and ~/.pi/agent/mcp.json by /gzmo setup.
The product path above is intentionally minimal: SQLite vault, embeddings off, no LAN services.
For the full curated-memory topology (operator / multi-host — honeypot at scale, graph, vectors):
Service | Typical role in GZMO |
SQLite | Durable vault + FTS (always) |
Redis | Hot session / cache plane |
Neo4j | Graph / relationship memory MCP |
Qdrant | Vector recall (when embeddings enabled) |
Use the full repo: maximilianwruhs-cyber/GZMO
Read:
docs/PRODUCT_MCP.md — product vs ops tools
Living / CT101 deploy docs in that repo (docs/CT101_DEPLOY.md, docs/PI_GZMO_MEMORY_INTEGRATION.md)
Deploy Redis + Neo4j + Qdrant next to your living gzmo.toml
On the operator machine, prefer scripts/install-shared-mcp.sh (LAN gzmo-memory + Neo4j) instead of the laptop product fragment
Keep this gzmo-pi package on workstations for the SQLite product vault; keep living MCP on the host that owns Redis/Neo4j/Qdrant
In Pi: /gzmo living prints the short pointer.
Command | Action |
| Init |
| Binary, config, MCP wiring checklist |
| Pointer to full Redis/Neo4j/Qdrant stack docs |
git clone https://github.com/maximilianwruhs-cyber/gzmo-pi.git
cd gzmo-pi
npm install
pi install .Publish (maintainers):
npm login # interactive; or: npm config set //registry.npmjs.org/:_authToken=…
npm pack --dry-run # expect ~tens of KB, not tens of MB
npm publish --access public
# then: https://pi.dev/packages/gzmo-pipi-mcp-adapter is a normal dependency (installed by pi install / npm install), not bundledDependencies — bundling it pulled ~57MB of transitive SDK tree into the tarball.
Product binary / vault: GZMO
Pi packages: pi.dev/packages
MCP adapter: pi-mcp-adapter
This server cannot be deployed
Personal knowledge graph as an AI memory layer over MCP - read, save, and link your memories.
Persistent personal memory for AI assistants — save, search, and recall across every MCP client.
Your memory, everywhere AI goes. Build knowledge once, access it via MCP anywhere.
Persistent AI memory shared across Claude, ChatGPT, coding agents, and compatible MCP clients.
