Hermes Atlas
Desktop apps, web UIs & dashboards

memo-agent

lxfu1/memo-agent

Terminal AI coding assistant built with TypeScript and Ink, described as a simplified Hermes Agent

In short

memo-agent is a standalone terminal AI assistant that its README calls a simplified version of Hermes Agent. It is written in TypeScript with React and Ink, connects to OpenAI-compatible APIs, and adds persistent notes memory, MCP tools and context compression.

What memo-agent does

memo-agent is a separate TypeScript application, not a plugin for Hermes. It borrows ideas from Hermes Agent and implements them in a terminal UI built with React and Ink. It connects directly to any OpenAI-compatible API, configured with MODEL_API_KEY, MODEL_BASE_URL and MODEL_NAME or a config file under ~/.memo-agent.

Persistent memory lives in a NOTES.md file that is injected into the system prompt each round and can be updated automatically at the end of a round. When the context grows, a three-zone compression archives the middle of the history and links the new session to the old one by parent_session_id. Other features include slash commands such as /notes, /history, /search, /compact, /cost and /plan, a plan, execute and reflect loop, custom recipes written as .md templates, MCP tool servers, a Brave Search web tool, SQLite session storage with /resume, and isolated profiles. A permission guard forces confirmation for dangerous commands, and an optional sandbox limits which environment variables child processes inherit.

Key features

  • NOTES.md persistent memory injected into the system prompt each round
  • Three-zone context compression with linked session chains
  • Slash commands including /notes, /history, /search, /compact, /cost and /plan
  • Plan, execute and reflect loop with tasks saved in SQLite
  • Custom recipes as markdown templates and MCP tool server support
  • Profiles, ask or auto permission modes and an optional RunCommand sandbox

When to use it

  • Trying a Hermes-style terminal assistant against any OpenAI-compatible endpoint
  • Keeping separate work and personal profiles with isolated memory and sessions
  • Studying how persistent notes and context compression can be implemented in TypeScript

Who it is for: Developers who want a lightweight Hermes-inspired terminal agent they can run with their own API key.

How it fits with Hermes Agent

A separate reimplementation described in its README as a simplified Hermes Agent, tagged hermes-agent. It does not run Hermes Agent itself.

How to install memo-agent

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

npm install -g memo-agent
export MODEL_API_KEY=sk-...
memo

Requirements: Node.js and an API key for an OpenAI-compatible model endpoint

Note: It is a standalone reimplementation inspired by Hermes Agent, not part of the Hermes ecosystem's runtime.

FAQ

What is memo-agent?

memo-agent is a terminal-based AI assistant built with TypeScript, React and Ink. Its README describes it as a simplified version of Hermes Agent with persistent memory, MCP tools and context compression.

Does memo-agent work with Hermes Agent?

No, it does not run on or integrate with Hermes Agent. It is a separate app that reimplements some Hermes-style ideas and talks to OpenAI-compatible model APIs directly.

How do I install memo-agent?

Run npm install -g memo-agent, set MODEL_API_KEY, optionally MODEL_BASE_URL and MODEL_NAME, then start it with memo. A config file at ~/.memo-agent/config.yaml is an alternative to environment variables.

Similar interfaces for Hermes Agent

All interfaces

Related guides: Connect Hermes agents on several machines with Hermes Desktop · How to install Hermes Agent