Stop Renting
Your AI's Memory
VMN is the open-source local node for xLMP — the AI Memory Control Plane. Deterministic, content-addressed local memory for AI agents. No cloud required. No embedding-model drift. Your data stays on your NVMe.
Context Collapse Is Killing
Your AI Workflows
Your retrieval results change whenever the embedding model updates, chunking strategy shifts, or similarity threshold is retuned. The same query returns different chunks six months later.
Your memory lives on someone else’s cloud. Every query is a billable API call. Changing providers means re-embedding your entire knowledge base.
Clinical data, trade secrets, and proprietary codebase context uploaded to external embedding servers. One breach exposes everything.
Three Commands.
Permanent Memory.
Feed any text to vmn_ingest. xLMP breaks it into 1800-character deterministic memory segments, SHA-256 addressed and stored on your local NVMe.
Query with vmn_recall. The xLMP lexical scorer evaluates keyword density across all memory segments and returns the exact bounded evidence window. No embeddings. No drift.
Upgrade any memory segment to a Sovereign Cryptographic Record. ExergyNet LNES-04 issues a RISC Zero receipt, settled on-chain via Groth16. Cryptographically verifiable against the committed memory. Cross-team shareable without exposing raw content.
Up in 60 Seconds
Then add to your Claude Desktop config:
Or run without installing: "command": "npx" with "args": ["-y", "vanguard-memory-node"] · WSL: "command": "wsl" with "args": ["-d", "Ubuntu", "--", "vanguard-memory-node"]
Deterministic vs Probabilistic Memory
| Feature | xLMP — VMN | Vector DB (Chroma / Pinecone) |
|---|---|---|
| Retrieval method | Lexical keyword density | Cosine similarity |
| Result consistency | Deterministic — same query, same result | May vary — model updates, chunking, or threshold changes alter results |
| Privacy | Local NVMe only | Cloud upload required |
| Cost | Free (local) | Per-query API cost |
| Setup time | 60 seconds | Hours (embed pipeline) |
| Cryptographic proof | Via LNES-04 (coming soon) | Not available |
* Vector retrieval results may change when embedding models, chunking strategies, indexes, or similarity thresholds change.
Open Core.
Sovereign Infrastructure.
Full VMN core — local ingest, xLMP recall, stdio MCP. No limits. No telemetry. Self-hosted, open source.
Shared memory vaults across a workspace. Managed sync, access controls, and usage dashboards via the ExergyNet portal.
ZK-verified memory records — RISC Zero receipt, Groth16-settled. For clinical, legal, and regulated-context deployments.
Private ExergyNet deployment, custom SLA, dedicated memory network node, and integration support for large-scale agent infrastructure.
VMN is open. The ExergyNet memory network and production control plane are commercial infrastructure.
Your memory. Your NVMe.
Your agents.
VMN puts xLMP memory on every developer's machine. ExergyNet connects those memories into the sovereign AI Memory Control Plane.