Hermes Agent

An open-source, autonomous AI agent framework by Nous Research with a persistent memory loop that creates and refines its own reusable skills over time.

Free (MIT-licensed) + your own LLM API costs, or an optional Nous Portal subscription ($20, $100 or $200/month) with credits for 300+ models

My take

Typical agents treat every task as a blank slate, which gets annoying fast. Hermes has a built-in learning loop, so it actually builds and saves reusable skills as it works. If it figures out how to refactor a file or parse a document once, it remembers that pattern for the next task. It's model-agnostic too, so you aren't locked into a single provider — for a free, self-hosted framework, that pairing is rare.

Visit Hermes Agent Back to tools

At a Glance

Pros

+ Built-in learning loop allows agent to create and refine reusable skills
+ Persistent memory of user projects, files, and preferences across sessions
+ Model agnostic — works with 200+ models via OpenRouter or local endpoints
+ Open-source and highly customizable
+ One gateway connects it to 20+ messaging platforms, including Telegram, WhatsApp, Signal, Slack, Discord, Teams and email

Cons

− Setup is easier than it was but still more technical than a consumer app, with API keys and model endpoints to manage
− Skill learning loop can occasionally save inefficient patterns
− You pay for LLM usage yourself, through your own API keys or a Nous Portal subscription

Best for: Developers seeking a model-agnostic agent framework that learns from experience and builds a persistent memory of user preferences.

Full review

The Hermes Agent is an open-source, autonomous AI agent framework developed by Nous Research. While most AI assistants start every conversation with a clean slate, Hermes is designed to overcome this limitation by building a persistent memory of user preferences, project context, and custom skills.

The self-improving skill loop

The core innovation in Hermes is its learning loop. When you give the agent a task, it doesn’t just execute the steps once and forget them. If it encounters a new challenge and successfully writes a script or parses a file, it saves that pattern as a reusable “skill.” Over time, the agent builds a library of custom tools tailored specifically to your codebase and projects, improving its efficiency and success rate on subsequent runs.

Model agnostic by design

Unlike proprietary agent platforms tied to a single model family, Hermes is completely agnostic. You can connect it to over 200 different large language models via API providers like OpenRouter or local endpoints using Ollama. This flexibility allows developers to switch models depending on the task — using a fast, cheap model for simple file operations and swapping to a larger, reasoning-heavy model for complex architecture changes.

Persistent project memory

Hermes maintains project history and user context across sessions. It remembers your preferences, coding conventions, and past decisions, reducing the amount of background information you have to provide with every new request.

Technical setup

Setup has become easier. Since June 2026 there’s a desktop app, Hermes Desktop, for macOS (Apple Silicon), Windows and Linux; its installer brings its own runtime, so Docker isn’t required. It still asks for more technical comfort than a consumer app: you set up API keys and configure model endpoints, or subscribe to the optional Nous Portal ($20, $100 or $200 a month, with credits for 300+ models) instead. The software itself stays free and MIT-licensed. It is a powerful workspace for power users but not designed for casual non-technical builders.

Who should use Hermes

Software engineers and product managers who want to build a customized, self-improving development assistant. The ability to switch between open-weight models and the persistent learning loop make it a compelling choice for teams building custom automation pipelines.