Nous Research has released Hermes Agent, an open-source autonomous AI agent built around a self-improvement loop that allows it to write, update and delete its own procedural skills as it learns from use, positioning it as a technically distinct alternative to the dominant OpenClaw framework.
The project has accumulated more than 5,000 GitHub stars since its release and prompted widespread comparison with OpenClaw in developer communities, though the two tools are built around fundamentally different philosophies.
Hermes ships with 47 built-in tools, a persistent memory system that carries knowledge across sessions, Model Context Protocol (MCP) server integration, and voice mode support across platforms.
Its headline capability is a skills system that saves successful workflows as reusable procedures, meaning the agent becomes progressively more capable the more it is used rather than starting each session from scratch.
The agent supports multiple AI models including Claude, ChatGPT, Qwen and local open-source alternatives, and runs on infrastructure ranging from a Raspberry Pi to a cloud cluster, with near-zero cost when idle on serverless deployments.
OpenClaw, backed by significant development resources following OpenAI's acquisition of its creator and ongoing Nvidia support, has become the default framework for many users through its extensive ecosystem: daily updates, native plugin support for development environments including Cursor and Claude Code, and a growing number of commercial implementations from companies including Xiaomi, MiniMax and Nvidia's own enterprise offering.
Where OpenClaw functions as a broad orchestration layer that users configure through community-built skills, Hermes begins with more built-in capability and develops a personalised, expanding skill set through interaction, reducing reliance on third-party skills that carry potential security risks including prompt injection.
The two agents are not mutually exclusive: Hermes supports the Agent Communication Protocol (ACP), allowing it to operate alongside OpenClaw in multi-agent configurations, with OpenClaw handling orchestration while Hermes executes individual tasks.
Both carry meaningful security considerations, however, and neither should be treated as safe by default in production environments without careful privilege configuration and review of third-party integrations.
Related reading
- Microsoft confirms 365 Copilot agent with OpenClaw-like local capabilities ahead of Build showcase
- Memento-Skills lets AI agents rewrite their own capabilities without retraining underlying models
- Microsoft warns AI agents risk becoming "double agents" as it unveils security controls at RSAC
Installation requires a single terminal command that handles all dependencies, and setup takes under five minutes with no Docker or configuration files required.
The release adds to a rapidly expanding field of autonomous agent frameworks, with developers increasingly running multiple agents in parallel across different models to maximise performance on complex, multi-step tasks.The recap
- Decrypt published an article examining Hermes and OpenClaw.
- Hermes is described as a self-improving artificial intelligence agent.
- Article frames Hermes as a competitor 'coming for OpenClaw'.
