← all repositories
NousResearch/hermes-agent

An AI agent that autonomously edits its own skill set

Hermes exists so you can stop re-teaching assistants that are supposed to remember, and run them on a $5 VPS instead of your laptop.

219k stars Python AgentsLLMOps · Eval
Feature · 11 Jun 2026
The Agent That Refuses to Forget: Why Nous Research Built a Memory-First AI

Hermes Agent treats persistent memory and autonomous skill creation as core architecture, not bolt-ons, in a field still figuring out what "agent" actually means.

Read the in-depth article
hermes-agent
Velocity · 7d
+500
★ / day
Trend
cooling
star history

What it does

Hermes is a terminal-first AI agent that chats, runs tools, and delegates work to isolated subagents. It connects to a wide range of LLM providers — from OpenAI and OpenRouter to Xiaomi MiMo and Kimi — through a single TUI or messaging gateway. The agent persists memory across sessions, autonomously creates skills after complex tasks, and runs scheduled jobs via a built-in cron scheduler.

The interesting bit

The unusual part is the closed learning loop: Hermes nudges itself to persist knowledge, searches its own conversation history with FTS5, and refines its skills during use rather than waiting for a human to update prompts. It also builds a dialectic user model via Honcho, so it theoretically grows a deepening model of who you are over time. That, plus the ability to hibernate on serverless backends like Modal and Daytona, makes it feel less like a chatbot and more like a long-term coworker that happens to run on a cheap VM.

Key highlights

  • Model-agnostic by design: switch providers with a single command, no code changes required
  • Cross-platform messaging gateway: Telegram, Discord, Slack, WhatsApp, Signal, and CLI from one process
  • Self-curating memory: agent-managed persistence with periodic nudges, cross-session search, and LLM summarization
  • Autonomous skill creation: generates and improves skills from experience, compatible with the agentskills.io standard
  • Serverless-friendly: supports Modal and Daytona backends that hibernate when idle, plus Docker, SSH, and local terminals

Caveats

  • Native Windows support is early beta and explicitly flagged as less road-tested than the Linux, macOS, and WSL2 paths
  • The full feature set on Android/Termux requires a curated install because some voice dependencies are incompatible with the platform
  • The browser-based dashboard chat pane currently requires WSL2 on Windows, even though the core CLI and gateway run natively

Verdict

Developers who want a persistent, cloud-native agent with genuine memory and skill evolution should look here; those who just need a quick, stateless chat wrapper should probably look elsewhere.

Frequently asked

What is NousResearch/hermes-agent?
Hermes exists so you can stop re-teaching assistants that are supposed to remember, and run them on a $5 VPS instead of your laptop.
Is hermes-agent open source?
Yes — NousResearch/hermes-agent is open source, released under the MIT license.
What language is hermes-agent written in?
NousResearch/hermes-agent is primarily written in Python.
How popular is hermes-agent?
NousResearch/hermes-agent has 219k stars on GitHub and is currently cooling off.
Where can I find hermes-agent?
NousResearch/hermes-agent is on GitHub at https://github.com/NousResearch/hermes-agent.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.