Memory Bank Plugin for OpenClaw and Hermes

What Memory Bank Plugin for OpenClaw and Hermes does

The integration stores extracted facts, not raw conversation logs, in a Memory Bank instance inside your own Google Cloud project, so there is no vector database or embeddings infrastructure to run. Memory is user-scoped and can be shared across agents when the project, location, reasoning engine and scope match. For OpenClaw it ships as a plugin with automatic recall before each turn, automatic capture after substantive turns, workspace memory-file sync and CLI commands.

For Hermes it is an MCP server over stdio with five user-invoked tools, including memorybank_search and memorybank_remember. The README is explicit that Hermes gets no native lifecycle auto-recall or auto-capture here, so the agent has to call those tools. Setup needs a Google Cloud project with billing and the Agent Platform API enabled, the IAM roles aiplatform.user and aiplatform.memoryUser, a Memory Bank instance created through the Vertex AI SDK, and Application Default Credentials.

Key features

• Managed long-term memory in your own Google Cloud project
• Hermes MCP server over stdio with five memory tools
• memorybank_search and memorybank_remember tools for explicit recall and capture
• OpenClaw plugin with automatic recall and capture
• User-scoped memory shareable across agents with matching configuration
• Memories stored as extracted facts instead of raw transcripts

When to use it

• Keep user preferences and decisions across Hermes sessions
• Share one memory store between an OpenClaw agent and a Hermes agent
• Avoid running a vector database for agent memory

Who it is for: Teams that already use Google Cloud and want managed long-term memory for Hermes Agent or OpenClaw.

How it fits with Hermes Agent

Hermes is one of the two supported agents; it connects through an MCP server, so recall and capture are explicit tool calls.

Requirements: A Google Cloud project with billing and the Agent Platform API enabled, the roles aiplatform.user and aiplatform.memoryUser, a Memory Bank instance, google-cloud-aiplatform 1.111.0 or newer, and Application Default Credentials.

Note: The README says this is not an officially supported Google product, and it needs a Google Cloud project with billing enabled.

FAQ