Hermes Agent has surpassed Open Claw to become the top open-source AI agent on Open Router, processing 224 billion tokens daily. The recent v0.13 'Tenacity' release introduces persistent multi-agent Kanban boards, background macOS computer use, 95% token compression, voice cloning, multilingual support, and robust security, making it a powerful, cost-effective, and versatile AI agent for developers and users worldwide.
Hermes Agent, an open-source AI agent developed by Nuke's Research Lab, has recently overtaken Open Claw to become the number one AI agent on Open Router, the global leaderboard for AI tools. Processing an astonishing 224 billion tokens per day, Hermes is now the most used open-source AI agent worldwide, marking a significant shift in the AI agent landscape.
Open Router functions like the Billboard chart for AI tools, ranking agents and applications based on their usage and popularity. Hermes Agent's rise to the top spot on this leaderboard is a testament to its growing adoption and capabilities.
Unlike most AI agents such as Cursor, Claude Code, Codex, Pod Code, or Kilo Code, which are session-based and forget previous interactions once closed, Hermes Agent is a persistent autonomous agent. It runs continuously in the background, building long-term memory from every conversation and learning from experience through a built-in learning loop.
Hermes Agent's learning loop consists of four key components:
This means Hermes gets smarter over time, unlike other agents that are powerful but amnesic.
Hermes is MIT licensed, fully open source, and can run on various hardware, from a $5 VPS to a high-end RTX 4090 rig. It supports over 20 messaging platforms including Telegram, Discord, Slack, WhatsApp, Signal, iMessage, and Microsoft Teams, allowing users to interact with it from anywhere without cloud lock-in or telemetry concerns.
On May 7th, Nuke's Research released Hermes Agent v0.13, dubbed the "Tenacity" release. This update focuses on reliability and completing tasks without interruption.
Kanban Multi-Agent System: The Kanban board now supports multiple agents working in parallel with features like heartbeat tracking, automatic task reclaim, zombie detection, retry budgets, and hallucination recovery. This allows tasks to be completed autonomously without user babysitting.
Goal Command: Upgraded from beta to a first-class feature, the goal command enables autonomous objective mode where Hermes plans, executes, reviews, and retries tasks until completion. It integrates tightly with the Kanban and survives restarts.
Checkpoints V2: Improved state pruning and disk space management prevent bloating and orphaned repositories.
Session Persistence: Conversations now survive gateway crashes, updates, and reloads, allowing seamless continuation.
Security Enhancements: Eight critical security issues were fixed, including cross-skill DM bypass, default redaction for debug logs, stranger rejection on WhatsApp, race condition fixes, and SSRF blocking on cloud metadata endpoints.
Hermes Agent introduces a unique computer use feature for macOS that allows it to operate in the background without taking over the user's machine. Unlike other implementations where the agent controls the cursor and keyboard focus, Hermes uses Apple's Skylight private SPIs and accessibility frameworks to send synthesized events directly to target applications.
This means users can continue working on their computer while Hermes performs tasks like browsing or clicking in other applications silently in the background.
You can set a goal such as monitoring your email, summarizing important messages, posting summaries to Telegram, and adding response tasks to the Kanban board. Hermes will perform these actions in the background without interrupting your workflow.
Currently, this feature is macOS only, with Linux and Windows support planned. It relies on private Apple APIs, which may break with OS updates, but version pinning is recommended for stability.
A standout feature of the v0.13 release is the four-layer context compression mechanism, which reduces token usage by approximately 95% during long-running sessions.
Long-running AI agent tasks typically consume large context windows, increasing costs exponentially. With this compression, running an 8-hour session that might have cost $50 can now cost as little as $2.50.
This makes persistent agents economically viable for hobbyists and developers using local open-source models with smaller context windows.
Qwen 3.6 Plus Support: Adds a 1 million token context window model supported by Alibaba, providing a strong alternative to GPT and Claude, especially for users outside the US.
Voice Cloning via XAI: Hermes supports custom voice cloning for text-to-speech, enabling personalized and accessible voice interactions.
Multilingual Support: CLI and gateway messages now support seven new languages including Chinese, Japanese, German, Spanish, French, Ukrainian, and Turkish.
Google Chat Integration: Hermes now supports Google Chat, expanding its presence to 20 messaging platforms.
Video Analysis Tool: Enables native video understanding with Gemini and other multimodal models.
Self-Linting Code: Hermes automatically checks its own code for syntax errors before execution, reducing cascading failures.
Autonomous Curator: The agent reviews and consolidates its own skills, archiving stale ones and maintaining an efficient skill set.
Cron Scheduling: Built-in automation allows scheduled tasks with delivery to any supported messaging platform.
Windows Support: Early beta available with PowerShell one-liner; WSL2 is recommended for stability.
Installing Hermes Agent is straightforward:
The entire process takes about 15 minutes.
Six months ago, the open-source AI agent space was fragmented and unstable, with most users preferring closed-source agents from big labs. Hermes Agent's rapid development and adoption demonstrate that open-source solutions are now competitive in real-world usage.
Its architecture—persistent memory, self-improving skills, multi-platform gateway, and model-agnostic design—sets the direction for the future of AI agents. Big labs are already adopting similar patterns, but Hermes is available today, free, and open source.
If you want an AI agent that respects your data, runs on your hardware, supports multiple models, learns over time, and reliably completes complex tasks, Hermes Agent is the one to explore. While not perfect and with some rough edges, its rapid development and feature set make it a standout in the AI agent ecosystem.
The "Tenacity" release and its groundbreaking features like background computer use, token compression, multi-agent Kanban, multilingual support, and voice cloning represent a new era for open-source AI agents. For developers, hobbyists, and teams, Hermes Agent offers a powerful, flexible, and cost-effective tool to build and run autonomous AI workflows.
Stay tuned for future updates, as the pace of innovation from Nuke's Research is extraordinary, promising even more capabilities in the near future.
