Nous Research’s Hermes Agent has overtaken OpenClaw to become the most-used open-source agent on OpenRouter’s daily inference rankings. The framework crossed 175,000 GitHub stars in four months and now processes over 220 billion tokens per day, according to a comprehensive field guide published by Marily Nika on June 8.
The shift happened in May 2026. ChatForest reported that Hermes hit #1 on OpenRouter’s global token rankings on May 6, processing 271 billion tokens across all users and surpassing OpenClaw’s 186 billion daily. Hermes had reached 140,000 GitHub stars in under three months at that point, faster than any prior open-source agent framework reached that milestone.
What Separates Hermes from the Field
Hermes launched in February 2026 with a specific architectural bet: agents should get smarter the longer you use them. The framework runs as a persistent process on infrastructure you control, from a $5/month VPS to serverless backends like Modal and Daytona that hibernate when idle.
Three design choices drive adoption. First, persistent memory: Hermes maintains two curated markdown files (MEMORY.md and USER.md) injected into every conversation, capped at roughly 1,300 tokens total. The constraint is intentional. As the Substack field guide explains, “unbounded memory is how you get an agent that’s expensive, slow, and weirdly fixated on something you said in March.”
Second, a self-improving skill system. After completing complex tasks, Hermes writes reusable skill documents stored in ~/.hermes/skills/. ChatForest’s analysis cites Nous Research benchmarks showing agents using self-created skills completed research tasks 40% faster than fresh instances, with the improvement compounding over subsequent runs.
Third, model agnosticism. Hermes connects to over 300 models through OpenRouter, including Claude, GPT, Gemini, DeepSeek, and Nous Research’s own Hermes 3 model built on Llama 3.1. Users switch models with a single command.
Platform Reach
Hermes connects to over 16 messaging platforms out of the box: Telegram, Slack, Discord, WhatsApp, Signal, Matrix, SMS, email, and others. This cross-platform gateway means users interact with their agent from any surface without opening a dedicated app. The agent also includes a built-in cron system configured in plain English for scheduled autonomous tasks.
The GitHub repository lists nearly 1,000 contributors. A desktop app for Mac and Windows shipped in May, wrapping the same agent core in a GUI.
The Competitive Picture
OpenClaw held the #1 position on OpenRouter’s rankings for months before Hermes overtook it. The two frameworks solve different problems at different layers: OpenClaw is a CLI-native agent harness focused on code and system operations, while Hermes targets persistent, cross-platform personal agents that accumulate knowledge over time.
The speed of the transition is the story. Hermes went from zero to the most-used open-source agent framework in under four months. For teams evaluating open-source agent infrastructure, the framework war now has a genuine second front.