Hermes Agentvs OpenClaw

OpenClaw is a free, MIT-licensed assistant from the OpenClaw Foundation. You run its Gateway on your own computer, a server or a VPS, and OpenClaw states it has no hosted service. Its FAQ says sleep and reboots interrupt a laptop Gateway and recommends a VPS for 24/7 reliability. Its docs link guides for third-party hosts such as Fly.io, Hetzner, GCP and Azure.

Self-hosted, Hermes Agent also needs a machine that stays on, because its gateway runs the scheduled jobs and chat connections. Its commands can execute locally, in Docker, over SSH, or in Modal, Daytona or Vercel Sandbox. Hermes Cloud is the hosted option: a dedicated, always-on instance with its own workspace and dashboard, in the nearest available region.

You chat with a Hermes Cloud agent in the browser or through messaging channels such as Telegram and Discord. Hermes Cloud manages the agent's version and keeps your data and setup when you stop the instance. Every paid Nous Portal plan includes one Hermes Cloud agent, up to the plan's included size; larger or extra agents bill daily from credits. Without a plan, a small credit balance is enough to deploy one.

Nous Portal is the recommended way to run Hermes Agent, in the words of the Hermes Agent docs. One login covers hundreds of models, including Claude, GPT, Gemini, DeepSeek, Qwen and Grok models, with Hermes Agent usage included in paid plans within limits that reset daily, weekly and monthly. Paid plans also include the Tool Gateway for web search, image generation, text-to-speech and cloud browser automation, so you need no separate account with a search, image or browser vendor.

If your OpenClaw setup uses several providers, hermes setup --portal collapses it to that one login. Hermes Agent also works with an Anthropic API key, any OpenAI-compatible API and local models through Ollama, vLLM, llama.cpp or SGLang. It offers ChatGPT sign-in for Codex models and signs in with SuperGrok or X Premium+. /model switches models mid-session, and backup providers are opt-in with hermes fallback.

OpenClaw states it has no paid tier, and you pay each model provider you use. It works with many providers and local models, signs in with ChatGPT for Codex models and with SuperGrok, and can reuse a Claude Code login on the same machine.

Bot Mode is built into Hermes Desktop and on by default. Each bot has its own role, model, memory, skills and avatar. Bots run recurring routines, work together in group chats of 2 to 6, and message each other directly. Since v0.23.0, a group chat can let a lead bot or an automatic router pick who answers, and scheduled jobs can post into a room.

One gateway owns every session, so the CLI, terminal UI, Hermes Desktop, bots and cron share one conversation. Hermes Desktop connects to every Hermes Agent backend you own, from the local runtime and remote gateways to SSH hosts and Hermes Cloud instances. Hermes Desktop runs on macOS (Apple Silicon), Windows and Linux. The optional Webapp, off by default, serves the same workspace in a browser tab from your own gateway.

OpenClaw hosts several isolated agents on one Gateway with routing rules, and its agents can reach each other's sessions and spawn sub-agents. Its broadcast groups let several agents reply to one message in a group chat, and its ACP sessions run coding tools such as Claude Code, Cursor, OpenCode or Gemini CLI.

Both agents learn as they work. Hermes Agent keeps two curated memory files, MEMORY.md and USER.md, and stores every CLI and messaging session in a database with full-text search. When it works out a non-trivial workflow, it saves the approach as a skill and can update it later. A background curator maintains the skills it creates.

By default, Hermes Agent writes skills and memory freely. Turn on the write-approval gate, and every skill write, from a live turn or the background review, is staged until you approve it. Memory writes have the same optional gate. Staged skill writes survive restarts, and you review them with /skills pending and /skills approve, in the CLI or from a messaging app.

OpenClaw's self-learning is on by default. The agent saves a procedure after hard multi-step work and fixes a learned skill that misled it. Every change applies immediately and can be undone, and an optional propose mode turns background reviews into drafts you review first. Learned skills nobody uses for 30 days are archived.

ClawHub is OpenClaw's public registry for skills and plugins. Hermes Agent's Skills Hub integrates ClawHub as a community source, alongside skills.sh, LobeHub, browse.sh, GitHub repositories and the official optional skills. You can search ClawHub from the CLI, or install from a ClawHub card in the public Skills Hub.

Hermes Agent's skills are compatible with the agentskills.io open standard. For tools, it adds plugins from a curated catalog of over 1,050 entries and MCP servers over stdio or HTTP. ClawHub also lists plugins, and those are built for OpenClaw; Hermes Agent uses ClawHub for skills.

Run hermes migrate openclaw. It installs the official agent-importers plugin when it is missing and runs the import. On a fresh install, hermes setup looks for ~/.openclaw (and legacy Clawdbot or Moltbot folders) and offers the same import. The import always shows a full preview first, --dry-run stops at the preview, and by default a restore-point backup of ~/.hermes is written before the import applies.

The import brings over your persona, memory files and daily notes, skills, default model and custom providers, MCP servers, text-to-speech settings, and messaging settings for Telegram, Discord, Slack, Signal, Matrix and Mattermost. API keys and bot tokens come over only with --migrate-secrets, which hermes setup passes for you. Settings with no direct equivalent, such as cron jobs and plugins, are archived for you to review.