Connect YantrikDB Cognitive Memory to Hermes via MCP

How to use

Memory lives in a local SQLite file (default ~/.yantrikdb/memory.db); nothing leaves your machine in stdio mode.

Use Hermes's native memory (MEMORY.md, USER.md, skills) as the default — it is the right tool for identity, user preferences, and procedural knowledge the agent actively curates.

Reach for YantrikDB in the following situations:

You are storing thousands of facts, not hundreds, and context-window cost matters.

You want automatic contradiction detection, such as 'Alice moved to Berlin' versus an older 'Alice lives in London.'

Temporal decay matters — stale memories should fade unless explicitly reinforced.

Multiple agents, devices, or sessions need to share a common memory (SSE mode).

For a single-user laptop setup, native memory is simpler and sufficient.

The YantrikDB engine is AGPL-3.0, while the MCP server wrapper is MIT. Using the engine as-is via the MCP server does not trigger AGPL obligations on your code, but modifying the engine and redistributing it does. If you are building a commercial product, read the license before forking.

This is not an official Nous integration: YantrikDB is a third-party project that works with Hermes because both speak MCP, so if the Hermes MCP client or the YantrikDB server changes shape, the bridge needs to hold on both sides.

Treat it as an add-on, not a replacement: Hermes's native memory handles identity and style, and YantrikDB is best as a searchable long-term fact store the agent can query on demand.

For a lighter alternative, Hermes's built-in skill_manage and session search (FTS5-backed) cover most 'remember this across sessions' needs without adding a service.

Stdio mode is for a single machine and is the simplest option:

pip install yantrikdb-mcp

Add the server to Hermes's MCP server config, typically config.yaml under mcp_servers::

mcp_servers:
  yantrikdb:
    command: yantrikdb-mcp
    # Optional: override default DB location
    env:
      YANTRIKDB_DB_PATH: /home/you/.hermes/yantrikdb.db

Restart Hermes. The YantrikDB tools now appear under hermes tools.

SSE mode shares memory across machines, for example Hermes on a VPS talking to YantrikDB on a homelab box:

Run the server in SSE mode so it listens for connections from other machines:

export YANTRIKDB_API_KEY=$(python -c "import secrets; print(secrets.token_urlsafe(32))")
yantrikdb-mcp --transport sse --port 8420

Then point Hermes's MCP config at the SSE endpoint:

mcp_servers:
  yantrikdb:
    type: sse
    url: http://your-homelab:8420/sse
    headers:
      Authorization: Bearer YOUR_API_KEY