When an AI agent can learn from every conversation, create its own skills, remember who you are across sessions, and talk to you from Telegram while simultaneously running code on a remote GPU cluster 鈥?you know we’ve crossed into a new era. That’s exactly what Hermes-Agent, the open-source creation of Nous Research, delivers. And it’s not just another AI assistant 鈥?it’s the first agent with a genuine built-in learning loop that gets smarter the more you use it.
With over 60,000 GitHub stars and a staggering 6,438 stars in a single day, Hermes-Agent has become one of the most talked-about projects in the AI open-source ecosystem. But the numbers only tell part of the story.
What Makes Hermes-Agent Different
Most AI agents today are essentially sophisticated API wrappers 鈥?they take your instructions, call some tools, and return results. They don’t really learn from their experiences. Close the chat, and they’re back to square one, with no memory of what happened before.
Hermes-Agent breaks this pattern entirely. Built by the team at Nous Research, it introduces what the developers call a closed learning loop 鈥?a system where the agent doesn’t just perform tasks but actively improves its approach based on what it learns while doing them.
Here’s what that looks like in practice:
- Agent-curated memory with periodic nudges: Hermes actively decides what to remember from each session and periodically reminds itself of important context
- Autonomous skill creation: After completing complex multi-step tasks, Hermes automatically creates reusable skills that can be applied to future problems
- Skills that self-improve during use: Each time a skill is invoked, the agent evaluates how well it worked and refines it
- FTS5 session search with LLM summarization: The agent can search through its own conversation history to recall relevant past experiences
- Honcho dialectic user modeling: A sophisticated system for building a deepening model of who you are, your preferences, and your working style
Run It Anywhere 鈥?From a VPS to a GPU Cluster
One of the most practical breakthroughs of Hermes-Agent is its deployment flexibility. This isn’t software that requires a rack of expensive GPUs to run. According to the project documentation, you can run it on a VPS, a GPU cluster, or serverless infrastructure 鈥?and the serverless option costs nearly nothing when idle.
The agent supports six different terminal backends: local, Docker, SSH, Daytona, Singularity, and Modal. Daytona and Modal deserve special mention because they offer true serverless persistence 鈥?the agent’s environment hibernates when you’re not using it and wakes up on demand when you send a message. This means you can have a personal AI agent running 24/7 on a cloud VM, accessible via Telegram, and it only actually costs compute when there’s work to do.
Use Any Model You Want
Unlike proprietary AI agents that lock you into a single provider, Hermes-Agent is model-agnostic by design. You can connect to:
- Nous Portal (the developers’ own platform)
- OpenRouter with access to 200+ models
- z.ai/GLM
- Kimi/Moonshot
- MiniMax
- OpenAI
- Anthropic
- Or your own custom endpoint
Switching models is as simple as running hermes model 鈥?no code changes, no configuration files to edit, no lock-in.
A Full Terminal Interface 鈥?and a Messaging Gateway
Hermes ships with a full terminal user interface featuring multiline editing, slash-command autocomplete, conversation history, interrupt-and-redirect, and streaming tool output. It feels like a professional development environment, not a chatbot.
But it doesn’t stop there. The same agent can also connect to your existing communication platforms through its messaging gateway:
- Telegram
- Discord
- Slack
- Signal
This means you can be having a conversation with your AI agent on Telegram, and it can simultaneously be working on a coding task on a remote server. When it’s done, it messages you with the results.
Built-In Scheduling and Automation
The agent also includes a built-in cron scheduler that can deliver reports to any platform. You can set up daily reports, nightly backups, weekly audits 鈥?all defined in natural language, running unattended. The agent delegates and parallelizes work by spawning isolated subagents for separate workstreams.
Research-Ready: RL Training and Trajectory Generation
For AI researchers, Hermes-Agent includes batch trajectory generation, Atropos RL environment integration, and trajectory compression for training the next generation of tool-calling models.
Migrating from OpenClaw
Interestingly, Hermes-Agent includes a migration path from OpenClaw. Running hermes claw migrate automatically detects your OpenClaw installation and offers to import your persona files, memories, skills, command allowlists, messaging settings, API keys, and TTS assets.
Get Started
Installing Hermes-Agent takes about two minutes:
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
hermes setup
The full documentation is available at hermes-agent.nousresearch.com/docs, and the project is released under the MIT license.
Whether you’re a developer looking for a powerful research agent, a power user who wants AI assistance across your devices, or a researcher studying how agents can learn and improve over time, Hermes-Agent is one of the most interesting open-source projects to emerge in 2026.