What is Hermes? Nous Research's self-improving open-source agent
Hermes is an MIT-licensed agent that lives on your server, writes its own skills, and works with any model, local ones included. What ships today, how it installs, and the OpenClaw relationship.
Hermes Agent is Nous Research’s open-source personal AI agent: MIT-licensed, Python, and pitched in their own words as “an autonomous agent that lives on your server, remembers what it learns, and gets more capable the longer it runs.” Its signature is a built-in learning loop: it creates skills from its own experience and improves them during use. It connects to 20+ chat platforms, works with any model provider, including local ones, and installs by CLI on Linux/macOS or as an official desktop app on Mac and Windows. One disambiguation before anything else: “Hermes” is also the name of Nous Research’s LLM model series. The agent and the models are separate products that share a brand.
Agent vs models: the naming trap
Nous Research has shipped Hermes-branded model weights since 2023: Hermes 3, Hermes 4 (405B/70B/14B, August 2025), Hermes 4.3, all open models with a strong local-inference story. Hermes Agent arrived later (the release log dates it February 2026): a harness, like OpenClaw or Claude Code, that uses models rather than being one. Nothing requires pairing them: the agent’s docs are explicitly model-agnostic. When someone says “Hermes” in an agent-hosting conversation, they usually mean the agent; when a benchmark thread says it, they mean the models.
What makes it distinctive
Every serious harness has tools, channels, and memory. The README leads with the part Hermes claims as unique: “the only agent with a built-in learning loop — it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and builds a deepening model of who you are across sessions.” Mechanically that means agent-authored skills (compatible with the agentskills.io open standard, shareable via a skills hub), cross-session recall with summarization, and user modeling. The rest of the sheet, per the docs: 60+ built-in tools (browser automation, vision, image generation, TTS, terminal, MCP integration), isolated subagents, a scheduler with delivery to any platform, and messaging across “20+ platforms” (Telegram, Discord, Slack, WhatsApp, Signal, email, SMS, Teams, Matrix, and more) from one gateway process that installs as a systemd/launchd service. It moves fast: v0.20.0 (August 2026) added streaming voice with barge-in and an agent-to-agent protocol. Three newer capabilities have grown into subjects of their own: Hermes kanban, the multi-agent task board that lets named profiles work a shared queue, first-class Agent Client Protocol support (hermes acp), which seats Hermes inside ACP editors and workspaces, and Bot Mode, which turns profiles into a roster of named Bots that message each other.
Models: bring anything, including local
Provider support covers Nous’s own portal, OpenRouter, OpenAI, Anthropic, Gemini, DeepSeek, and any OpenAI-compatible endpoint, which is the local door: point it at Ollama, vLLM, llama.cpp server, or SGLang and the docs accommodate local inference explicitly (they even stretch the response timeout to 30 minutes when a local endpoint is detected, because local prefill on big contexts is slow). Note what that is and isn’t: the project documents local-model support thoroughly; how well small local models hold up through long agentic runs is something practitioners debate, not something the docs promise. The software itself is free: “you pay only for the LLM API usage from your chosen provider.”
Installing and running it
Official paths: a CLI installer for Linux, macOS, WSL2, and Termux (install.sh piped to bash, the distribution model the whole agent generation shares, with the same verify-what-you-run caveat), a PowerShell installer for native Windows, Docker images, and Hermes Desktop apps for macOS 12+ (Apple Silicon) and Windows 10/11, but no Linux desktop app. Like OpenClaw, it’s an always-on gateway process at heart (hermes gateway start installs the background service), so the where-does-it-live question is the same one: a machine that stays on. The README’s own suggestion list starts at “a $5 VPS.” Hardware minimums aren’t documented; budget it like an OpenClaw-class Node/Python service rather than a model host, unless you’re serving local models on the same box, which changes the math entirely.
The OpenClaw relationship
Hermes and OpenClaw are the two most-watched open-source personal agents, and Hermes courts switchers directly: an official hermes claw migrate command reads ~/.openclaw and imports settings, memories, skills, and API keys. The docs also note one workflow difference for movers: where OpenClaw binds multiple independent agents to specific chats, Hermes runs one agent across your surfaces (with profiles for isolated instances). For the philosophical comparison (architecture, security posture, which community fits you), our OpenClaw vs Hermes guide goes deep. The short version: OpenClaw is the configure-everything power tool with the bigger ecosystem; Hermes bets on the agent improving itself. Both assume what this whole category assumes: a real computer of the agent’s own, running around the clock.