Hermes Agent, the open-source personal agent from the AI research lab Nous Research, is back on GitHub Trending — and its pitch is unusual even in a year full of agent launches: it says it gets better the more you use it, because it learns from its own experience.
The project, which has become one of the most popular open-source agent efforts of 2026, describes itself as 'the agent that grows with you.' At the core is what Nous calls a closed learning loop. After a complex task, the agent can autonomously create a skill; those skills are improved while they are in use. The system nudges itself to persist knowledge, can search its own past conversations (with FTS5 indexing and LLM summarization), and builds a deepening model of the user across sessions. It is compatible with the emerging agentskills.io open standard for agent skills.
Nous Research is known in the AI community for the open Hermes model family and for its work on open-weight AI. Hermes Agent extends that line from models to the software around them: a self-hosted, MIT-licensed agent that works with whatever model you choose — Nous Portal, OpenRouter, OpenAI, or a custom endpoint — with no lock-in. Switching providers is a single 'hermes model' command.
It is also built to live outside the laptop. A full terminal UI is complemented by a messaging gateway that lets users talk to the agent from Telegram, Discord, Slack, WhatsApp, Signal or email, including voice-note transcription. A built-in cron scheduler runs automations such as daily reports or nightly backups in natural language. For bigger jobs, the agent can spawn isolated sub-agents for parallel workstreams and call Python tools via RPC.
The infrastructure story is deliberately flexible: seven terminal backends — local, Docker, SSH, Singularity, Modal, Daytona, Vercel Sandbox — mean it runs on a $5 VPS, a GPU cluster, or serverless setups that hibernate when idle and wake on demand. Native Windows support arrived without requiring WSL.
There is a research angle too. Hermes can generate and compress batch trajectories, which Nous positions as useful for training the next generation of tool-calling models — a reminder that open agents double as data-generating infrastructure.
The project also shows how the open-agent economy is consolidating. Nous now sells 'Nous Portal', a single subscription covering 300+ models plus a tool gateway (web search, image generation, text-to-speech, cloud browser). Hermes also ships an automatic migration path from OpenClaw, importing settings, memories, skills and API keys.
Skeptics still ask whether 'self-improving' agents are a real category or a marketing label for elaborate memory systems. For now, the GitHub star count is answering for the community: the project is adding roughly 575 stars a day as it climbs the trending charts.
