1. Origin and Developer OpenClaw was developed by Austrian Peter Steinberger, published in November 2025 under the name Clawdbot. Hermes Agent was released by Nous Research on February 25, 2026.
2. GitHub Traction OpenClaw has accumulated 346,000 GitHub stars, becoming the fastest-growing open-source project in GitHub history. Hermes has surpassed 140,000 stars in less than three months.
3. Design Philosophy OpenClaw is an "agentic harness" platform: not an AI model, but a framework that gives agents the ability to act in the real world, connecting to any external LLM. Hermes is designed explicitly for long-term autonomous cloud deployment — "independent of local devices" is a design principle, not an afterthought.
4. Self-Improvement Here lies the most significant difference: after every task with 5 or more tool calls, Hermes executes a reflection phase and generates a reusable skill file. A background process called "Curator" re-evaluates and rewrites underperforming skills every week. OpenClaw uses a skill system but does not generate them autonomously.
5. Memory OpenClaw maintains persistent memory through Markdown files and SQLite, combining vector search and FTS5 to recall preferences and contexts. Hermes' memory architecture is more sophisticated: long-term semantic memory, working memory, and episodic logs work together for genuine recall between sessions.
6. Supported Messaging Platforms OpenClaw supports over 20 channels: WhatsApp, Telegram, Discord, Slack, Signal, iMessage, WeChat, Teams, and many others. Hermes supports Telegram, Discord, Slack, WhatsApp, Signal, and CLI — a more curated set, but expanding.
7. Security OpenClaw has received criticism for its broad permissions: it can access emails, calendars, system files. The Cisco team found a third-party skill that executed data exfiltration without the user's knowledge. Hermes adopts a more conservative approach: Nous Research curates and stress-tests every included skill and plugin, aiming to make the framework work with 30 billion parameter models without continuous debugging.
8. Supported LLM Models Both are model-agnostic. OpenClaw supports Claude, GPT, DeepSeek, Gemini, and local models via Ollama. Hermes adds Nous Portal, OpenRouter (200+ models), NovitaAI, NVIDIA NIM, MiniMax, Hugging Face, and any custom endpoint.
9. Ecosystem and Enterprise Adoption OpenClaw has had a huge impact in China: Tencent and other cloud providers have released services based on OpenClaw, and local governments have offered grants to startups building on it. Hermes already has a one-click template on Tencent Cloud Lighthouse, the first cloud provider to offer direct deployment of an open-source agent at this level.
10. Current Market Position On May 10, 2026, Hermes surpassed OpenClaw as the most used agent on OpenRouter, processing 224 billion tokens daily compared to OpenClaw's 186 billion. However, OpenClaw still holds the cumulative historical lead (9.17 trillion tokens vs 6.35 trillion).
Final Summary OpenClaw is the pioneer: it created the category of open-source local AI agent frameworks, with a massive community, global virality, and the highest number of users. It's flexible, hackable, and designed for those who want full control over their stack — but this openness exposes it to non-trivial security risks. Hermes is the evolved response: it comes from a serious research lab (Nous Research) and is built with a different thesis — an agent that actively improves itself, preferably in the cloud, with a more curated and reliable architecture. It pays a price in terms of community size and virality but is already technically surpassing OpenClaw in daily usage. In summary: OpenClaw is the entry point and cultural phenomenon; Hermes is where the technology is headed.