Hermes Agent (Nous Research) and OpenClaw are the two most popular open-source personal AI agents. Hermes is built around a learning loop that writes its own skills; OpenClaw is built around a gateway that puts one assistant in 30+ chat apps and on your devices. Here is a sourced, side-by-side comparison and a simple way to pick.
Checked October 6, 2026, against both projects' GitHub repositories and official docs. Star counts and versions change weekly; re-check before you decide.
Pick Hermes Agent if you want one agent that learns your work and gets better at it. Pick OpenClaw if you want one assistant that lives in every chat app and device you use, with the biggest skill ecosystem. Both are free, MIT-licensed, self-hosted personal AI agents that work with any major model. Hermes, from Nous Research, is built around a learning loop that writes and refines its own skills. OpenClaw, from the OpenClaw Foundation, is built around a Gateway that connects one assistant to 30+ messaging channels and companion apps.
| Category | Hermes Agent | OpenClaw |
|---|---|---|
| Maker | Nous Research | OpenClaw Foundation (independent 501(c)(3)); created by Peter Steinberger |
| License | MIT | MIT |
| Architecture | Python agent with TUI, a messaging gateway, and seven terminal backends (local, Docker, SSH, Singularity, Modal, Daytona, Vercel Sandbox) | Node.js Gateway as the local control plane; Control UI, CLI, TUI, and companion apps for macOS, iOS, Android, Windows, Linux |
| Memory | MEMORY.md + USER.md (small, curated), SQLite full-text search of past sessions, optional memory plugins | USER.md, MEMORY.md, daily notes, hybrid vector + keyword search, background "dreaming" consolidation |
| Skills | Creates and updates its own skills after tasks; optional human approval gate; agentskills.io standard | SKILL.md (AgentSkills spec); agent drafts skill proposals for operator approval; ClawHub public registry |
| Channels | 29 platforms in the gateway docs (Telegram, Slack, WhatsApp, Teams, Signal, email, SMS and more) | 34 channels in the docs, plus WebChat and device nodes (voice, camera, screen) |
| Models | Any: Nous Portal, OpenRouter, OpenAI, your own endpoint; switch with hermes model | Hosted and local providers; Claude, Codex, and local models plug in as harnesses |
| Install | One-line installer (macOS, Linux, WSL2, native Windows), Android via Termux, Hermes Desktop app | One-line installer or npm (Node 24.16+), Docker, Nix; onboarding wizard |
| Security defaults | Command approval (smart/manual/off), hard-blocked destructive commands, DM pairing, hardened containers | DM pairing by default; tools run on the host for the main session unless you configure sandboxing |
| Software cost | Free; optional Nous Portal plans ($0, $20, $100, $200/mo) | Free |
| Best for | Developers and solo operators with repeat workflows | Multi-channel personal assistants, households, small teams, device control |
Sources: Hermes Agent README, Hermes messaging docs, OpenClaw README, OpenClaw channels docs, Nous Portal plans, all checked October 6, 2026.
Hermes Agent is an open-source, self-improving AI agent built by Nous Research. Its README calls it "the agent that grows with you" and describes a "built-in learning loop": it creates skills from experience, improves them during use, searches its own past conversations, and builds a model of who you are across sessions (GitHub).
A few facts worth knowing before you install it:
hermes claw migrate imports OpenClaw settings, memories, skills, and API keys, with a dry-run preview and a preset that skips secrets.OpenClaw is an open-source AI assistant that runs on your own computer and meets you in the chat apps you already use. Its README lists Discord, iMessage, Slack, Teams, Telegram, WhatsApp, and 20+ more, plus native apps for macOS, iOS, Android, Windows, and Linux (GitHub). We cover the basics in What is OpenClaw?
This is the real difference between the two projects.
Hermes keeps memory deliberately small. Two files, MEMORY.md (2,200 characters) and USER.md (1,375 characters), are injected into the prompt at session start. Everything else lives in a SQLite database with full-text search that the agent queries when it needs to recall an old conversation (Hermes memory docs). That keeps every session lean.
OpenClaw keeps memory broad. It writes USER.md, MEMORY.md, and dated daily notes in your workspace, searches them with hybrid vector and keyword search, flushes key facts to disk before a long conversation is compacted, and runs a background "dreaming" pass that promotes daily notes into long-term memory (OpenClaw memory docs).
Hermes writes its own skills. A skill_manage tool lets the agent create, update, and delete skills, which the docs call its "procedural memory". A background review can stage skill changes after a session, and you can require human approval before they land (Hermes skills docs).
OpenClaw proposes skills, then you approve. Its Skill Workshop drafts a proposal when the agent spots reusable work, instead of writing straight to SKILL.md. Its bigger advantage is ClawHub, the public skills registry, which now shows VirusTotal and static-analysis scan results before install (OpenClaw skills docs).
In short: Hermes compounds what you teach it; OpenClaw gives you more of what everyone else has built. DeepLearning.AI's comparison reached the same conclusion and noted that some users find Hermes less token-efficient (The Batch, May 2026).
Both agents can run shell commands, read files, and send messages for you, so security is about defaults and track record.
OpenClaw's record is public and mixed. CVE-2026-25253 affected versions before 2026.1.29: OpenClaw took a gatewayUrl value from a query string and connected to it automatically, sending a token without prompting (NVD). In February 2026, researchers at Koi Security found 341 malicious ClawHub skills, 335 of them from one campaign (ClawHavoc) that installed the Atomic Stealer macOS malware (The Hacker News). The project has since added scan results to ClawHub, and its README still warns that tools run on the host for the main session unless you configure sandboxing.
Hermes ships stricter defaults. Its docs describe eight security layers: smart command approval (an auxiliary model rates risk; uncertain cases go to you), a hard block on catastrophic commands even in "yolo" mode, DM pairing codes that expire in one hour, and Docker containers run with all capabilities dropped (Hermes security docs). Fewer public incidents also reflects a younger, smaller skill ecosystem, not immunity.
See our OpenClaw vs Dooza breakdown for what a self-hosted setup asks of you.
Both are free software. Your bill is the model you run, the machine it runs on, and any paid tools (search, voice, browser).
| Cost line | Hermes Agent | OpenClaw |
|---|---|---|
| Software license | Free (MIT) | Free (MIT) |
| Model access | Bring your own keys, or Nous Portal: Free $0, Plus $20/mo, Super $100/mo, Ultra $200/mo (credits included) | Bring your own keys or local models |
| Hosting | Laptop, VPS, or idle-hibernating serverless (Modal, Daytona) | Your computer, a server, or Docker |
| Hidden cost | Your time on setup, updates, and approvals | Same, plus vetting third-party skills |
Nous Portal prices are from portal.nousresearch.com, checked October 6, 2026. Nous pitches the Portal as the simplest way to power Hermes:
Hermes is the agent that grows with you.
— Nous Research (@NousResearch) July 20, 2026
Nous Portal is the best way to power it: bundling the models, useful external tools, and the cloud agent itself under a single unified subscription with incredibly simple setup.https://t.co/4KFwhsReyA pic.twitter.com/cQOvpL0LXa
Claude Code is a coding agent; Hermes is a general personal agent. Anthropic describes Claude Code as "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools," available in the terminal, IDE, desktop app, and browser. Most surfaces need a Claude subscription or Anthropic Console account (Claude Code docs).
| If you want to… | Pick | Why |
|---|---|---|
| Automate the same research, report, or ops task every week | Hermes | Self-written skills and cron jobs compound over time |
| Reach your assistant from iMessage, WhatsApp, Teams, and your phone | OpenClaw | Most channels plus native mobile and desktop apps |
| Run on a cheap cloud box that sleeps when idle | Hermes | Modal and Daytona backends hibernate between sessions |
| Install a ready-made skill for a popular app today | OpenClaw | ClawHub is the largest public registry (read skills before enabling) |
| Share one assistant across a small team or household | OpenClaw | Documented team deployments on one Gateway |
| Try the other one without starting over | Hermes | hermes claw migrate imports OpenClaw memories and skills |
| Have AI answer customers, chase leads, or post content for a business, without running a server | Neither, self-hosted | A managed service such as Dooza Workforce fits better |
For the wider field, see OpenClaw alternatives, and for the terminology, AI agents vs agentic AI.
Tech With Tim set up both agents on real servers and ran the same jobs through each in a July 2026 video. His video description frames the stakes:
Six months ago everyone was running OpenClaw. Now Hermes Agent has surpassed it — 224 billion tokens per day vs OpenClaw's 186 billion.
The chapters worth jumping to are "The Key Difference - Skills" (3:30), "Hermes Agent Curator" (12:01), and "Security Issues". Treat usage figures as a snapshot: DeepLearning.AI reported in May 2026 that Hermes had moved ahead of OpenClaw on OpenRouter's daily-token leaderboard, and those rankings shift week to week.
The installers take minutes. Running an agent safely for months is the real job. Before you connect either one to business accounts, plan for:
For a developer, this is a fun weekend. For a business owner, it is a part-time job.
Dooza is an AI-native company that builds AI products and services for small businesses, from the Dooza Workforce app to the Dooza Agents platform. Every product starts with a refundable pilot: 100% refund within 14 days.
Dooza is a managed alternative to self-hosting Hermes or OpenClaw. It is not built on either one. If you want the outcome (emails answered, leads followed up, calls picked up, content posted) without choosing a framework, patching a gateway, or vetting skills, there are two ways in:
Either way you get encrypted connections and your approval on anything sensitive, and no server to babysit. Pricing depends on the product; see pricing. Every Dooza product starts with a refundable pilot — 100% refund within 14 days.
The honest trade-off: if you enjoy tinkering, want full control of your data on your own hardware, or are building agents yourself, Hermes or OpenClaw is the better choice. If you want the work done, a managed option is faster.
Book a free pilot call and a Dooza engineer will scope your pilot in 30 minutes.
Both are free, MIT-licensed, self-hosted AI agents. Hermes Agent, from Nous Research, centers on a learning loop that writes and improves its own skills and keeps a small, curated memory. OpenClaw, from the OpenClaw Foundation, centers on a Gateway that connects one assistant to 30+ chat channels and device apps, with the largest public skill registry, ClawHub.
Neither is better for everyone. Hermes is better for developers and solo operators who repeat tasks and want the agent to learn them, and it ships stricter security defaults. OpenClaw is better when you need more channels, native mobile apps, team deployments, or ready-made community skills.
Hermes Agent is free and open source under the MIT license. You pay for the model and hosting you choose. Nous Research also sells an optional Nous Portal subscription, listed at $0, $20, $100, and $200 per month on October 6, 2026, which bundles model access and hosted tools.
Yes. Hermes includes a hermes claw migrate command that detects an existing OpenClaw folder and imports settings, memories, skills, and API keys. It supports a dry run and a preset that migrates everything except secrets.
Use Claude Code for software development: it reads your codebase, edits files, runs commands, and opens pull requests. Use Hermes Agent for a general, always-on personal assistant that works with any model provider across chat apps. Many developers use both.
Yes. Dooza is a managed alternative: Dooza Workforce provides ready-made AI employees, and Dooza Agents are custom agents built and maintained by Dooza engineers. Dooza is not built on Hermes or OpenClaw, and every Dooza product starts with a refundable pilot with a 100% refund within 14 days.
Both are free, MIT-licensed, self-hosted AI agents. Hermes Agent, from Nous Research, centers on a learning loop that writes and improves its own skills and keeps a small, curated memory. OpenClaw, from the OpenClaw Foundation, centers on a Gateway that connects one assistant to 30+ chat channels and device apps, with the largest public skill registry, ClawHub.
Neither is better for everyone. Hermes is better for developers and solo operators who repeat tasks and want the agent to learn them, and it ships stricter security defaults. OpenClaw is better when you need more channels, native mobile apps, team deployments, or ready-made community skills.
Hermes Agent is free and open source under the MIT license. You pay for the model and hosting you choose. Nous Research also sells an optional Nous Portal subscription, listed at $0, $20, $100, and $200 per month on October 6, 2026, which bundles model access and hosted tools.
Yes. Hermes includes a hermes claw migrate command that detects an existing OpenClaw folder and imports settings, memories, skills, and API keys. It supports a dry run and a preset that migrates everything except secrets.
Use Claude Code for software development: it reads your codebase, edits files, runs commands, and opens pull requests. Use Hermes Agent for a general, always-on personal assistant that works with any model provider across chat apps. Many developers use both.
Yes. Dooza is a managed alternative: Dooza Workforce provides ready-made AI employees, and Dooza Agents are custom agents built and maintained by Dooza engineers. Dooza is not built on Hermes or OpenClaw, and every Dooza product starts with a refundable pilot with a 100% refund within 14 days.
Automate your business with AI employees that work 24/7. Start with a refundable pilot: 100% refund within 14 days.
OpenClaw is free, MIT-licensed, and the most-starred personal agent on GitHub, but you run it, pay the model bill, and own the security. Here are the alternatives worth a look in 2026, with sourced facts, a comparison table, and an honest note on when to stay with OpenClaw.
RingCentral AI Receptionist (AIR) is a voice AI agent that answers calls, handles FAQs, books appointments and routes callers. Here is what it costs per minute, how setup works, what real admins report, and when an alternative fits better.
Start with a refundable pilot — 100% refund within 14 days. A Dooza engineer scopes it with you on a free 30-minute call. Pricing depends on the product; see pricing.
Refundable pilot · 100% refund within 14 days · No contracts