Open-source AI agents · checked 2 October 2026
Run Hermes Agent if you want an agent that does scheduled work on its own and writes its own skills as it goes. Run OpenClaw if you want an always-on assistant that lives in 20+ chat apps and on your phone. Both are free, MIT-licensed and work with whatever model you bring. Each one can import the other's memory, so picking wrong costs you an afternoon, not your setup.
Short answer
My stack uses Hermes for the boring, recurring work: morning ideas, signup digests, health checks and decision logs that run on a schedule and only ping me when something needs me. OpenClaw sits on the browser-work side of my stack. If you only install one and your goal is “an agent that runs part of my business while I sleep,” start with Hermes.
If your goal is “an assistant I text from iMessage, Teams or WhatsApp and that also runs on my phone,” OpenClaw has the wider reach out of the box.
Side by side
Everything here comes from each project's own README and docs (sources at the bottom). Star counts move daily.
| What you care about | Hermes Agent | OpenClaw |
|---|---|---|
| Who builds it | Nous Research. | The OpenClaw Foundation, an independent 501(c)(3). Originally built by Peter Steinberger and the community. |
| License and price | MIT. The software is free; you pay for whatever model you plug in (or an optional Nous Portal subscription). | MIT. The software is free; you pay for whatever model provider you plug in. |
| Language | Python. | TypeScript on Node.js (Node 24.16+ or 26.1+ if you install via npm). |
| Install | One-line installer for Linux, macOS and WSL2, a PowerShell installer for native Windows, and an APT repo for Android/Termux. Then run hermes. | One-line installer for macOS, Linux and WSL2, a PowerShell installer for Windows, or npm install -g openclaw. Onboarding sets up the Gateway. |
| How it runs | A terminal app (TUI) you start when you need it. Run hermes gateway when you want it reachable from chat apps. | A Gateway that runs as a background service and owns sessions, tools and channels. The Control UI, CLI and TUI all connect to it. |
| Chat apps | Telegram, Discord, Slack, WhatsApp, Signal and Email from one gateway process, plus Home Assistant. | Discord, iMessage, Slack, Teams, Telegram, WhatsApp and 20+ more, plus native apps for macOS, iOS, Android, Windows and Linux. |
| Models | Any provider: Nous Portal, OpenRouter, OpenAI, Anthropic, or self-hosted endpoints like Ollama and vLLM. Switch with hermes model. | Hosted and local providers. The docs cover Anthropic, OpenAI, OpenRouter, Ollama, LM Studio, vLLM and many more. |
| Memory | Two small curated files (MEMORY.md, 2,200 chars; USER.md, 1,375 chars) loaded at session start, plus full-text search over every past session in SQLite. Optional external memory providers. | Plain Markdown in the workspace: MEMORY.md, USER.md and daily notes. Semantic memory_search over notes, and a background “dreaming” pass that promotes useful notes into MEMORY.md. |
| Skills | Creates skills from experience after complex tasks and improves them while using them. Follows the agentskills.io open standard. | Skills and plugins you install. ClawHub is the public place to share them. |
| Scheduled jobs | Built-in cron scheduler, written in plain English, with delivery to any connected platform. | Built-in cron jobs and scheduled tasks. |
| Where it can run | Seven terminal backends: local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox. Modal and Daytona can hibernate when idle. | Your laptop, a VPS, or a shared team deployment. Companion apps and nodes add voice, camera, screen and device actions. |
| Switching | hermes claw migrate imports your OpenClaw persona, memories, skills, allowlist, messaging settings and API keys. | openclaw migrate hermes imports from a Hermes install. |
| GitHub stars (2 October 2026) | 250,682 | 391,191 |
Hermes Agent
Hermes Agent is Nous Research's open-source AI agent. The pitch is a built-in learning loop: it creates skills after complex tasks, improves those skills while using them, keeps a small curated memory about you, and can search every past conversation.
You talk to it in a terminal app, or run its gateway and message it from Telegram, Discord, Slack, WhatsApp, Signal or Email. It has a built-in cron scheduler, so “every weekday at 8am, summarize X and send it to Telegram” is a plain-English job, not a script.
The memory design is deliberately small. MEMORY.md is capped at 2,200 characters and USER.md at 1,375, and both get loaded into the prompt at the start of each session. Anything bigger goes into skills or gets found later through session search. One practical tip straight from the docs: on chat apps, a conversation is one long session until you send /new, so reset at natural breaks or new memories won't show up.
That cron scheduler is why I use it. The Hermes desktop files are the exact profile config and cron blueprint I share so an agent runs itself instead of waiting for you to type.
OpenClaw
OpenClaw is an open-source personal AI assistant that runs on your own machine and meets you in the chat apps you already use. A local Gateway handles sessions, tools and channel connections. You manage it from a Control UI, CLI or TUI, and native companion apps on macOS, iOS, Android, Windows and Linux add things like voice, camera and screen actions.
Memory is plain Markdown you can open and edit: MEMORY.md for durable facts, USER.md for preferences, and dated daily notes. A background “dreaming” pass decides which notes graduate into long-term memory, and memory_search finds notes by meaning, not just exact words.
One thing to read before you connect it to anything public: per OpenClaw's own README, tools run on the host for the main session unless you configure sandboxing. Unknown senders on DM channels have to be paired first by default.
How to choose
Switching
You don't lose your setup either way. On Hermes, hermes claw migrate --dry-run previews what it will pull from ~/.openclaw: your SOUL.md persona, memories, skills, command allowlist, messaging settings and allowlisted API keys. The setup wizard offers this automatically if it detects an OpenClaw install.
Going the other way, openclaw migrate hermes --dry-run does the same from a Hermes install. If you run both side by side, give each agent its own bot token so they aren't fighting over the same messages.
Start in 15 minutes
My free Hermes Agent starter pack gives you 7 cron blueprints, 3 agent skills, 3 personas and my annotated config, so you get a self-running Hermes agent set up in about 15 minutes instead of staring at an empty config.yaml.
If you're still deciding between hosted “super agents” and running your own, read my Manus vs Genspark comparison first. And if you want the bigger picture of how I build a company around agents like these, that's what The 38X AI Company is about.
FAQ
It depends on the job. Hermes Agent is the better fit for scheduled, unattended work and for an agent that builds its own skills over time. OpenClaw is the better fit for an always-on personal assistant that lives in many chat apps and on your phone and desktop. Both are free and MIT-licensed.
No. Hermes Agent is a Python project from Nous Research. OpenClaw is a TypeScript project run by the OpenClaw Foundation. They are separate codebases, but each one ships an importer for the other.
Yes. Hermes Agent is open source under the MIT license. You still pay for the model you connect, such as OpenRouter, OpenAI or Anthropic, unless you run a local model. Nous Portal is an optional subscription that bundles models and tools under one account.
Yes. Run hermes claw migrate (add --dry-run to preview). It imports your SOUL.md persona, MEMORY.md and USER.md entries, user-created skills, command allowlist, messaging settings and allowlisted API keys. The setup wizard also offers to migrate when it finds ~/.openclaw.
Yes. Hermes supports self-hosted endpoints like Ollama and vLLM, and OpenClaw's provider docs cover Ollama, LM Studio and vLLM. Hermes' own docs warn that small local models (roughly under 30B parameters) often claim to save memory without actually calling the memory tool.
Hermes Agent supports Telegram, Discord, Slack, WhatsApp, Signal and Email from a single gateway. OpenClaw supports Discord, iMessage, Slack, Teams, Telegram, WhatsApp and more than 20 other channels.
Sources
Checked against the official repos and docs on 2 October 2026. Both projects ship fast, so check the docs for the latest.
Want the system behind these experiments? Read The 38X AI Company, my book on building AI work that learns.