Tech • IA • Crypto
Hermes Agent Just Broke OpenClaw With a New Insane Update
8/10TL;DR
Hermes Agent has rapidly overtaken OpenClaw in daily usage rankings, signaling a shift toward self-improving, memory-driven AI agents.
KEY POINTS
Hermes Surpasses OpenClaw in Daily Usage
On May 10, 2026, Hermes claimed the top spot on OpenRouter’s global daily rankings, processing roughly 224 billion tokens per day, compared to OpenClaw’s 186 billion. While token volume can be noisy, such rankings strongly influence developer attention and adoption in a fast-moving market.
Explosive Growth in Under 90 Days
Launched in February 2026 by Noose Research, Hermes has accumulated about 147,000 GitHub stars and 23,000 forks within months. By contrast, OpenClaw still leads historically with over 370,000 stars and 9 trillion cumulative tokens, but Hermes’ rapid rise highlights a shift in momentum rather than total dominance.
A Self-Improving Agent Model
Hermes is built around a “do, learn, improve” loop, where completed tasks are analyzed and converted into reusable skill files. This allows the system to accumulate procedural knowledge over time, moving beyond traditional assistants that reset with each new task.
Layered Memory System
The agent uses a three-tier memory structure: session memory for current tasks, episodic memory for past interactions, and procedural memory through learned skills. This enables long-term recall and pattern reuse without requiring heavy external infrastructure.
Local-First, Model-Agnostic Design
Hermes can run on local machines, servers, or cloud setups while remaining compatible with multiple providers, including OpenAI, Anthropic, AWS Bedrock, Nvidia NIM, and local models. This flexibility helps control costs and reduces dependence on a single vendor.
Major ‘Tenacity’ Update Accelerates Adoption
The v0.13 “Tenacity” release introduced 864 commits and contributions from 295 developers in one week. Key features include a multi-agent task board with monitoring, retry controls, and failure recovery, aimed at stabilizing long-running workflows.
Improved Task Coordination and Persistence
New capabilities such as the /goal command help maintain alignment across extended operations, addressing a common issue where agents lose track of objectives during complex, multi-step tasks.
Expansion Into an AI Operating System
Hermes is evolving toward an AIOS model, combining agent functionality with a visual interface (Ion UI) that allows users to monitor actions, files, and decisions. This transparency is increasingly seen as essential for trust and control in autonomous systems.
Aggressive Migration Strategy
Hermes includes tools to detect and import existing OpenClaw setups, including skills, memory, and configurations. This reduces switching friction and directly targets experienced users exploring alternatives.
Security and Reliability in Focus
Both platforms face challenges typical of powerful agent systems. Hermes has patched multiple vulnerabilities in recent updates, while OpenClaw has reportedly dealt with issues such as exposed instances and malicious repository entries, underscoring ongoing risks in the space.
Shifting Developer Priorities
Early sentiment suggests some users are drawn to Hermes for its ease of setup, adaptive memory, and learning capabilities, though migration estimates remain uncertain. The broader trend points toward agents that improve over time rather than simply integrate widely.
CONCLUSION
Hermes’ rapid ascent reflects a growing demand for AI agents that learn from experience and compound value over time, challenging established players and reshaping expectations for the next generation of intelligent systems.