ReMe: memory tool for Hermes Agent

ReMe

agentscope-ai/ReMe

Local-first Markdown memory kit for agents, with a Hermes Agent memory provider

In short

ReMe is a memory management kit that stores agent memory as readable Markdown files and retrieves it with BM25, optional embeddings and wikilink expansion. It includes a Hermes Agent memory provider that recalls before model calls and captures memory after completed turns.

What ReMe does

ReMe turns conversations and resources into Markdown memory with frontmatter and wikilinks, so users and agents can inspect, edit, move, sync and back up the files with ordinary tools. Indexes and generated metadata can be rebuilt. Conversations are progressively distilled into daily notes and long-term knowledge, and recall combines BM25, optional embeddings and wikilink expansion to return line-level passages without loading the whole knowledge base into context.

The Hermes Agent memory provider offers HTTP or embedded mode, automatic recall before model calls, asynchronous auto_memory after completed turns and profile-aware background work, and the README says it supports Hermes Agent 0.21 and later. The same workspace can be shared with other agents, including QwenPaw, DeepSeek Harness and OpenClaw, through native integrations, SKILL.md, CLI, HTTP, MCP or Python APIs. A browser-based ReMe Studio Playground lets you explore example memory files and linked memory graphs without installing anything.

Key features

Memory stored as Markdown files with frontmatter and wikilinks
Hermes Agent memory provider in HTTP or embedded mode
Automatic recall before model calls and asynchronous auto_memory after completed turns
Hybrid recall with BM25, optional embeddings and wikilink expansion
Shared workspace across agents through SKILL.md, CLI, HTTP, MCP or Python
Plugins for OpenClaw and DeepSeek Harness, and a browser Studio Playground

When to use it

Give a Hermes Agent profile durable, editable Markdown memory
Share one memory workspace between Hermes Agent and another agent
Browse and edit linked memory in the browser Studio Playground

Who it is for: Developers who want agent memory kept as plain Markdown files they can read, edit and back up.

How it fits with Hermes Agent

Provides a Hermes Agent memory provider (the repository is tagged hermes-plugin) alongside integrations for OpenClaw and DeepSeek Harness.

Requirements: Hermes Agent 0.21 or later for the Hermes integration

FAQ

What is ReMe?

ReMe is a local-first memory management kit for AI agents. It stores memory as editable Markdown files and builds a self-evolving personal knowledge base from conversations and resources.

Does ReMe work with Hermes Agent?

Yes, ReMe has a Hermes Agent memory provider with HTTP or embedded mode, automatic recall before model calls and asynchronous auto_memory after completed turns. The README says it supports Hermes Agent 0.21 and later.

Is ReMe free and open source?

Yes, the repository is licensed under Apache-2.0.