If every AI agent you’ve tried starts from scratch in each session, this is going to interest you.
Hermes Agent, from Nous Research, does something almost no other agent in the ecosystem does: it gets smarter the more you use it. Not in the abstract sense of “use the context from this conversation.” It literally builds skills from your sessions, refines them over time, and remembers who you are between conversations. The tagline is “The agent that grows with you” — and for once, it’s not just marketing.
Launched in February 2026, it reached 82,000+ stars on GitHub in less than two months. Here’s why so many developers are migrating.
What is Hermes Agent?
Hermes is an open-source CLI agent from Nous Research — the same lab behind the Hermes model series on HuggingFace (33 million cumulative downloads). The agent is completely independent of those models: it works with any model provider you choose.
Essentially, Hermes is a terminal agent that:
- Learns from your sessions and automatically builds reusable skills
- Runs on your machine, a $5 VPS, Docker, or serverless infrastructure
- Talks to you via Telegram, Discord, Slack, WhatsApp, Signal, or email
- Works with any LLM: OpenRouter (200+ models), OpenAI, your own endpoint, local Ollama
The key difference from tools like Claude Code or Cursor: those live in your editor, session by session. Hermes lives wherever you deploy it, stays online, and accumulates knowledge over time.
The Learning Loop: The Feature That Actually Matters
Most agents treat each conversation as a blank slate. Hermes has a built-in learning loop that works like this:
Session 1: You ask Hermes to deploy a FastAPI app to your VPS.
→ It executes it, observes what worked, creates a skill: "deploy-fastapi-vps"
Session 2 (next week): You ask for the same thing.
→ Hermes finds the existing skill, executes it, improves it based on what changed.
Session N: The skill is refined, tested, and reusable — by you or your team.
It’s not magic — it’s structured persistence. Hermes stores skills using the open standard agentskills.io (human-readable Markdown), so you can inspect, edit, and version everything it learns.
Compare that to OpenClaw, where skills are written manually. Hermes generates them from experience and then improves them on its own.
Installation in One Command
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
Works on Linux, macOS, WSL2, and Android via Termux. Native Windows is not supported — if you’re on Windows, install WSL2 first and run the command above.
The installer handles Python, Node.js, and all dependencies. The only prerequisite is having git.
After installing, run the setup wizard:
hermes setup
This guides you through three steps:
- Model provider: You choose between Nous Portal, OpenRouter, OpenAI, or any OpenAI-compatible endpoint. You can change it anytime with
hermes model— no config file editing. - Messaging gateway (optional): You connect Telegram, Discord, Slack, WhatsApp, Signal, or Email so Hermes is accessible from your phone while it works.
- Tools: You enable the toolsets you want. Hermes comes with 40+ built-in tools covering terminal, file system, web, code execution, and more.
hermes model # Change model provider anytime
hermes tools # Enable/disable tools
hermes gateway # Start the messaging gateway
hermes doctor # Diagnose any issues
Your First Conversation
Once configured, you launch the interactive CLI:
hermes
Try something practical to understand how it works:
You: Analyze this Python project and tell me what tests are missing.
Hermes:
→ Reads your project structure
→ Identifies existing test coverage
→ Lists gaps with specific recommendations
→ Asks if you want it to generate the missing tests
→ [Creates a skill: "analyze-missing-tests" for reuse later]
The second time you ask for something similar, Hermes finds the skill it created and executes it directly — faster, and already adapted to your project’s patterns.
Running Hermes on a $5 VPS
This is where Hermes differs from editor-based agents. You can deploy it on the cheapest VPS you find and have it available 24/7 from your phone.
# On your VPS (Ubuntu 22.04):
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
hermes setup
# Start the gateway so it's accessible from Telegram:
hermes gateway
# Or run it as a systemd service:
systemctl --user start hermes-gateway
systemctl --user enable hermes-gateway
Once the gateway is running:
- You send a message via Telegram from your phone
- It executes on the VPS, using your configured LLM provider
- It replies via Telegram
It’s a genuinely different workflow from Claude Code or Cursor — those require you to be at your computer. Hermes can be working while you’re away.
Cost note: With OpenRouter + an economical model for routine tasks and a premium model (Claude, GPT-4, Gemini) for complex reasoning, daily costs can stay very low. Hermes supports provider chains with fallback, so you can configure it to use the most economical model that can solve the task. The combination of VPS + any model + accumulated learning means infrastructure costs stay near zero while the agent keeps getting more capable — operational efficiency that compounds over time.
Coming from OpenClaw?
If you used OpenClaw, Hermes has an official migration path:
hermes claw migrate # Interactive migration
hermes claw migrate --dry-run # Preview what would migrate first
hermes claw migrate --preset user-data # Migrate without secrets
The setup wizard automatically detects ~/.openclaw during first configuration and offers to migrate before you set anything else up.
What migrates: settings, memories, skills, API keys, SOUL.md persona configuration, messaging platform tokens.
What to verify after: skills that reference OpenClaw-specific APIs need manual rewriting for Hermes’ tool system. Run hermes doctor to detect any issues.
Key Commands
hermes # Start interactive CLI session
hermes setup # Run the setup wizard
hermes model # Choose/change model provider
hermes tools # Configure enabled tools
hermes gateway # Start messaging gateway
hermes update # Update to latest version
hermes doctor # Diagnose issues
hermes claw migrate # Migrate from OpenClaw
Is It For You?
Hermes doesn’t replace Claude Code or Cursor if you want deep editor integration and AI assistance while you code. It’s a different category: a persistent, self-improving agent that lives in your infrastructure.
Use Hermes if:
- You want an agent that retains knowledge and improves between sessions
- You need it available 24/7, not just when your laptop is open
- You want total freedom with models — no vendor lock-in, switch providers whenever you want
- You’re migrating from OpenClaw or want a more cost-effective alternative
- You want to run it on affordable serverless infrastructure (Daytona, Modal) or a VPS
Stick with Claude Code/Cursor if:
- You want deep IDE integration and real-time AI assistance while coding
- Your workflow is primarily session-by-session, not long-running autonomous tasks
Get Started Now
# Install
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
# Configure
hermes setup
# Launch
hermes
Full documentation: Hermes Agent Documentation | Hermes Agent
GitHub (82k+ stars):
Have you tried it already? Are you running it on a VPS or locally? Let us know how you’re using it 