Hermes Agent, built by Nous Research, is drawing renewed attention from developers after packaging a one-click migration path for users leaving OpenClaw. The project first appeared on February 25 as version v0.1.0 and initially saw limited traction. Interest picked up after OpenClaw recorded 138 security vulnerabilities over 63 days, pushing parts of the developer community to reassess whether it was suitable for production use.
By early March, Hermes Agent had entered GitHub Trending and climbed as high as No. 11, with its star count passing 2,200 during that stretch. According to the source material, the repository now stands at 69,900 stars and 9,000 forks. AwesomeAgents described it as the “most ambitious open-source Agent launch of 2026 so far.”
Memory files and generated skills sit at the center
Nous Research presents Hermes Agent as a self-evolving AI agent rather than a simple coding assistant. Its design centers on a persistent memory system and a skill system that turns prior work into reusable assets. Two files carry much of that structure: MEMORY.md stores environmental details, conventions, and lessons from earlier tasks, while USER.md keeps user preferences and communication style. Both are injected into the system prompt at the start of each session.
Session history is also stored in a SQLite full-text search database, allowing the agent to retrieve conversations from weeks earlier. On the skills side, Hermes Agent automatically creates structured Markdown skill files after completing complex tasks, usually those involving more than five tool calls. Those files capture procedures, known information, and validation methods for later reuse. To reduce token use, Hermes loads only names and descriptions by default and fetches full skill content only when needed.
More than 40 built-in tools and six backend options
The platform includes more than 40 tools, covering web search, browser automation, visual understanding, image generation, and text-to-speech. It also supports natural-language scheduling, which lets the agent run recurring work such as report generation, data backup, and system monitoring without direct supervision. The article highlights several popular ecosystem components, including Hindsight, Anthropic-Cybersecurity-Skills, mission-control, Hermes Agent Self-Evolution, and Hermes Workspace.
Hermes Agent can also spawn independent sub-agents. Each sub-agent has its own conversation context, terminal, and Python RPC scripts, which allows parallel task pipelines without sharing context overhead. On infrastructure, the project supports six terminal backends: local execution, Docker, remote SSH, Daytona serverless, Singularity containers, and Modal cloud functions. The source says it can run on a $5 VPS or on GPU clusters, with commands sent through Telegram.
The split with OpenClaw is in the learning loop
The article frames Hermes Agent and OpenClaw as direct competitors in the open-source developer-agent category, though their architectures differ sharply. OpenClaw is described as being built around a centralized control plane that manages sessions, routing, tool execution, and state. Hermes Agent instead puts the agent’s own execution loop first, with gateways, schedulers, and runtimes arranged around an iterative cycle of doing, learning, and improving.
The gap is most visible in how each project handles skills. OpenClaw mainly relies on manually written skills loaded from different layers such as workspace, personal, shared, or plugin sources. Hermes Agent takes a different route: it has the agent generate skill files from experience. That closed learning loop is presented as its defining feature, along with a model of operation in which the system lives on a server, retains what it learns, and becomes more capable over time.
One-line install and guided setup lower the barrier
Installation is handled through a single command, with support for Linux, macOS, and WSL2. After setup, users can run hermes setup to launch a guided configuration flow, choose a model provider, connect messaging platforms such as Telegram, Discord, Slack, or WhatsApp, and start the first conversation. The article also lists common commands including hermes, hermes model, hermes tools, hermes gateway, hermes claw migrate, hermes update, and hermes doctor.
In terms of use cases, Hermes Agent is aimed at workflows that need persistent context across sessions, gradual skill accumulation, tool orchestration through browsers or Shell, and deployment across local hardware, cloud virtual machines, or low-cost serverless infrastructure. The source also points to team collaboration through Slack or Discord, cloud task execution triggered from Telegram, and trajectory export for generating RL training data for future tool-calling models.