Hermes Agent from Nous Research is one of the most popular open-source autonomous agents of 2026: MIT-licensed, around 248,000 GitHub stars, with persistent memory and a learning loop that turns experience into reusable skills. The agent itself is free — you only pay for the model behind it. This guide shows how to make that model free too.
What Hermes Agent does
- Learns as it works. It creates skills from experience, improves them during use, and searches its own past conversations.
- Lives where you are. An interactive CLI (
hermes) plus a messaging gateway (hermes gateway) for Telegram, Discord, Slack, WhatsApp, Signal, and email. - Runs tools anywhere. Terminal backends include local, Docker, SSH, and several cloud sandboxes, and a built-in scheduler handles recurring automations.
- Uses any model. Nous Portal, OpenRouter, OpenAI, 70+ providers in total, any OpenAI-compatible endpoint, and local runners such as Ollama, vLLM, llama.cpp, and LM Studio.
Install
Linux, macOS, WSL2, and Termux:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashWindows (PowerShell):
iex (irm https://hermes-agent.nousresearch.com/install.ps1)Reload your shell, then run hermes setup for the configuration wizard or just hermes to start chatting. Settings live in ~/.hermes/ (or %LOCALAPPDATA%\hermes\ on Windows).
Connect a free model
Run hermes model to switch providers. Three free options work well:
1. Nous Portal's $0 plan
The most native option: Nous Portal includes :free models with standard rate limits on its free plan. Use hermes setup --portal and pick a free model.
2. OpenRouter or NVIDIA Build
Choose OpenRouter in hermes model, paste your key, and select a free model such as z-ai/glm-5.2:free or nvidia/nemotron-3-super-120b-a12b:free. For the strongest free models — GLM 5.3 and Kimi K3 — add NVIDIA Build as a custom OpenAI-compatible endpoint with base URL https://integrate.api.nvidia.com/v1.
3. A local model
For fully private, unlimited use, point Hermes at a local server: Ollama (http://localhost:11434/v1) or LM Studio (http://localhost:1234/v1). Pick a model with solid tool calling; small models under 10B parameters tend to fail multi-step tasks.
Which free model to pick
| Need | Model | Why |
|---|---|---|
| Best overall agent | GLM 5.3 | Top free AA Intelligence Index (44.8), tool calling, 1M context |
| Hard reasoning | Kimi K3 | 94% GPQA Diamond, image input |
| Many cheap steps | GLM 5.3 Flash | Fast, free at three providers |
| OpenRouter only | Nemotron 3 Super | Built for agents, free :free route |
The live list is on the free models for AI agents page.
Pitfalls
- Agents burn requests. One task can take dozens of calls. OpenRouter's 50 requests a day (1,000 after a one-time $10 top-up) disappears fast; NVIDIA Build and Nous Portal are better for sustained use.
- Memory holds your data. Hermes stores memories and skills locally. Combined with free endpoints that may log prompts, avoid feeding it secrets.
- Give it guardrails. Start with the Docker terminal backend rather than your host shell until you trust your setup.
Looking for alternatives? Compare Hermes with OpenClaw, OpenHands, Agent Zero, and others in our free AI agents list.