Memory (local) logo

Memory (local)

Beta

The reference knowledge-graph memory server. It saves what the agent chooses to remember to a local file.

  • Productivity
  • Local
Runs
On the developer's Mac
Launched with
npx, Docker
Platform
macOS
Publisher
Model Context Protocol

How it works

Every agent request passes through Ferentin policy before it reaches Memory (local).

It runs on the developer's Mac and talks to the AI client over stdio, so no network gateway sees it. Developers launch it with npx -y @modelcontextprotocol/server-memory or docker run -i --rm mcp/memory.

Ferentin's endpoint agent finds Memory (local) in the MCP configuration of AI clients on managed Macs. An administrator adds it once from the local MCP catalog for every managed Mac that launches it this way. Each of its tools starts unvetted until it is approved. Ferentin records every tool call.

  1. AI agentAsks for something
  2. Ferentin policies
  3. Memory (local) MCP serverRuns on the developer's Mac
  4. Governed resultChecked, logged, returned

About Memory (local)

The reference knowledge-graph memory server. It saves what the agent chooses to remember to a local file.

Frequently asked about Memory (local)

How do I connect Memory (local) to Ferentin?
Enroll your Macs with the Ferentin endpoint agent, then add Memory (local) from the local MCP catalog in the Ferentin admin console. Every managed Mac that launches it the way the catalog names it is covered, and each of its tools starts unvetted until you approve it.
Is the Memory (local) integration generally available?
The Memory (local) integration is currently beta on Ferentin.