Hermes Agent from Nous Research: an open source AI with an integrated self-improvement loop
This MIT-licensed framework was launched in February 2026 and garnered 22,000 GitHub stars and 242 contributors in just a few weeks. This pace reflects a particular enthusiasm within the open source AI community. Users want agents that remember, adapt, and improve without requiring constant manual instructions.
Hermes Agent is built around what Nous Research calls a closed learning loop. When the agent solves a task, it writes a reusable Markdown skills file, stores the result in persistent memory, and adjusts its approach for the next time. No YAML. No manual configuration. The system handles all of this internally using SQLite full-text search and large-scale language model (LLM) summarization.
Openclaw, the AI agent framework that experienced stunning success in early 2026 with hundreds of thousands of stars on GitHub, works according to a different logic. It connects LLMs to existing chat applications (WhatsApp, Telegram, Slack, Signal, Discord, iMessage) and performs concrete actions: email management, browser automation, shell commands and cron jobs.
It’s fast, versatile, and community-driven. Its Clawhub skills marketplace is vast. Its limitation, as users have publicly pointed out, is that memory between sessions requires manual configuration. Hermes was designed to fill this gap.
Its persistent memory spans sessions, platforms and devices. It parses the inputs to inject prompts. It builds a user profile over time. Give it a research task just once, and it completes it, converts the process into a skill, and executes faster the next time it encounters a similar problem.
Installation is done in a single command. Migration from Openclaw is integrated: hermes claw migrate extracts skills, memories and settings from an existing Openclaw configuration. The migration cost is low. Many users have adopted it.
Hermes supports over 200 models via Nous Portal, Openrouter, OpenAI compatible endpoints and local Ollama. It includes over 40 built-in tools covering terminal access, file operations, browser automation, code execution, image generation, and natural language scheduling. It runs on a $5 VPS, locally, in Docker or via SSH.
It was developed by Nous Research. The lab, led by co-founder and CEO Jeffrey Quesnelle, focuses on open source model development and reinforcement learning. Mr. Quesnelle demonstrated that the agent was capable of independently writing a 79,000-word novel during iterative sessions. The project Discord is active. The skills hub on agentskills.io continues to grow.
Security was directly addressed in v0.5, which Nous Research labeled a hardening release. It included over 200 pull requests, fixed a LiteLLM credential exposure issue, added path traversal fixes, and improved container security. Built-in security measures include order approval flows, blocking unsafe templates, and memory scanning for injections. No reports of widespread violations exist in public records.
The community consensus on Reddit, YouTube and X is not that Hermes replaces Openclaw. Most users consider them complementary. Openclaw manages multi-channel operations, team workflows and ecosystem depth. Hermes manages the learning layer: persistent memory, automatically generated skills, pattern reasoning, and long-term tasks. A common configuration is to run Hermes as a high-level scheduler on top of the Openclaw tools.
Quesnelle and contributors continue to release versions. Hermes v0.6.0 was released at the end of March 2026 with support for MCP and SDK-typed workflows. Deposits move quickly. For those already using Openclaw, the migration only takes a few minutes. What happens next depends on your usage level. The learning loop intensifies. That’s the whole point.
FAQ 🤖
- What is Hermes Agent? Hermes Agent is a free, MIT-licensed standalone AI framework developed by Nous Research that learns from completed tasks and builds persistent memory across sessions and platforms.
- How is Hermes Agent different from Openclaw? Openclaw focuses on orchestration and diversity of plugins in chat applications, while Hermes Agent automatically generates reusable skills and retains long-term user memory without manual configuration.
- Is Hermes Agent safe to use? Hermes Agent includes order approval flows, dangerous template blocking, and sandboxed container support, with active security patches released in its hardened v0.5 release.
- Can I migrate from Openclaw to Hermes Agent? Yes, running the “hermes claw migrate” command automatically transfers your existing Openclaw skills, memories, and settings to the Hermes Agent.
