Hermes Agent vs OpenClaw: how they differ, and using both

Updated September 28, 2026

Hermes Agent and OpenClaw are both open-source, MIT-licensed personal agents that you run on your own machine or server and talk to from a terminal or a chat app. Hermes Agent, from Nous Research, is built around a learning loop: it creates and improves its own skills and searches its past sessions. OpenClaw, from the OpenClaw Foundation, is built around one Gateway that connects your chat channels to one or more isolated agents.

Who makes Hermes Agent and OpenClaw

Hermes Agent is built by Nous Research and released under the MIT license. Nous Research also offers Nous Portal, an optional subscription for models and tools, but Hermes works with whichever provider you choose.

OpenClaw is stewarded by the OpenClaw Foundation, an independent 501(c)(3), and is also MIT-licensed. It was built by Peter Steinberger and the community. The project says it has no paid tier, hosted service or token.

How you install and run each one

Hermes installs with one command on Linux, macOS and WSL2, with a PowerShell one-liner for Windows and a Termux package for Android:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Run hermes for the terminal interface, or hermes gateway to reach it from chat apps. Its tools can run in seven terminal backends: local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox. So the agent can work on a cloud VM while you talk to it from Telegram.

OpenClaw installs the same way on macOS, Linux and WSL2, with a PowerShell script for Windows, or with npm if you manage Node.js:

curl -fsSL https://openclaw.ai/install.sh | bash

Onboarding sets up the Gateway, the local control plane for sessions, tools and channels. You reach it from the Control UI, the CLI, a TUI, chat channels, and companion apps for desktop and mobile. Tools run on the host for the main session unless you configure sandboxing.

Which chat apps each one supports

Hermes’s gateway is one background process for every platform you configure, including Telegram, Discord, Slack, WhatsApp, Signal, SMS, Email, Matrix, Microsoft Teams, BlueBubbles (iMessage), Feishu/Lark and LINE. OpenClaw’s Gateway connects to Discord, iMessage, Slack, Teams, Telegram, WhatsApp, Google Chat, Signal and more than 20 others. If you rely on a less common app, check each list.

Models and providers

You pick a Hermes provider with hermes model: Nous Portal, OpenRouter, OpenAI, Anthropic, Google, GitHub Copilot, self-hosted endpoints such as Ollama and vLLM, and any OpenAI-compatible endpoint. It also supports fallback providers.

OpenClaw works with hosted and local providers. It ships provider plugins such as OpenAI and Anthropic, and accepts custom providers and local runtimes such as LM Studio, Ollama and vLLM. It treats models and agent harnesses (Claude, Codex, local models) as plugins you can swap.

How memory works in each

Both keep memory as plain files the agent writes, and both load a MEMORY.md and a USER.md when a session starts.

Hermes keeps these two short files in the profile’s memories folder, with character limits, and stores every session in SQLite with full-text search that the agent queries with session_search. Details: Hermes Agent memory.
OpenClaw keeps its memory in the agent’s workspace, adds daily notes, finds notes with a semantic memory_search, and runs a background “dreaming” sweep that promotes items into MEMORY.md. Details: OpenClaw memory.

In both, memory belongs to one profile or one agent.

Skills

Hermes skills are knowledge documents that load only when needed and follow the agentskills.io open standard. Hermes creates skills after complex tasks and improves them as it uses them, and you can add more from the Skills Hub.

OpenClaw skills are markdown instructions: each skill is a folder with a SKILL.md file and YAML frontmatter. OpenClaw loads bundled skills plus your own, filtered by environment, config and installed binaries. ClawHub is its public skills registry, and the Skill Workshop lets you review and approve skills the agent drafts.

Running several agents: Hermes profiles vs OpenClaw multi-agent

A Hermes profile is a separate Hermes home with its own config, .env, memory, sessions, skills, cron jobs and gateway state. hermes profile create coder makes one, and coder becomes its own command, as in coder chat or coder gateway start. The docs say never to point two agents at one profile, and to use an external memory provider when agents need shared memory.

OpenClaw runs several isolated agents inside one Gateway process. Each has its own workspace, state directory and session store. Bindings map a channel account, such as a Slack workspace or a WhatsApp number, to an agent, so inbound messages reach the right one. Skills can be shared across agents or allowlisted per agent.

So Hermes keeps each agent in its own home, with its own gateway, while OpenClaw runs them side by side in one Gateway.

MCP support: adding a remote server

Both connect to MCP servers over stdio or HTTP, and both can serve their chat conversations to other agents over MCP (hermes mcp serve, openclaw mcp serve).

In Hermes, you add a remote server under mcp_servers in config.yaml, with a url and headers, or with hermes mcp add. Servers that use OAuth take auth: oauth. In a running session, /reload-mcp loads the change.

In OpenClaw, saved servers go under mcp.servers in openclaw.json, or you run openclaw mcp add with --url and --transport streamable-http. Streamable HTTP servers need transport: "streamable-http". openclaw mcp doctor <name> --probe tests one.

Scheduling and background work

Hermes has a built-in cron scheduler. You create jobs in plain language or with cron expressions, and the agent manages them itself through its cronjob_manage tool. Jobs can attach skills, fire on webhooks, and deliver results to the chat they came from, to files or to other platforms.

OpenClaw’s scheduler is called Automations. It persists jobs, wakes the agent at the set time, and delivers output to a chat channel, a webhook or nowhere. You manage it with openclaw automations (openclaw cron also works), and it takes webhook and Gmail PubSub triggers.

Hermes Agent vs OpenClaw at a glance

	Hermes Agent	OpenClaw
Made by	Nous Research	OpenClaw Foundation, a 501(c)(3)
Installs on	Linux, macOS, WSL2, Windows, Android (Termux)	macOS, Linux, WSL2, Windows
Where tools run	Seven terminal backends, local to serverless	The host, or a sandbox you configure
Chat	One gateway process for every platform you configure	One Gateway, 20+ channels, companion apps
Models	Many providers via `hermes model`, local included	Hosted and local provider plugins
Memory	`MEMORY.md`, `USER.md`, session search	Workspace Markdown, daily notes, semantic search
Skills	agentskills.io standard, agent-created, Skills Hub	`SKILL.md` folders, ClawHub, Skill Workshop
Several agents	Profiles, one home each	Isolated agents in one Gateway, routed by bindings
Remote MCP	`mcp_servers` in `config.yaml`	`mcp.servers` in `openclaw.json`
Scheduling	Cron, managed by the agent	Automations

How many people use each

No single count covers both fairly. OpenRouter’s Personal Agents ranking by tokens, read on September 28, 2026 for “Today”, put Hermes Agent first with 1.73 trillion tokens and OpenClaw third with 177 billion. It only counts apps that route through OpenRouter, so it misses anyone running either agent on a provider key directly.

On GitHub, OpenClaw had 389,434 stars on September 11, 2026, according to onewave-ai.com. We found no current star count for Hermes, so we do not compare stars.

Which should you pick

Think of it as fit, not a winner.

Hermes Agent fits if you want an agent that writes its own skills over time, want its tools to run in Docker, over SSH or in a serverless sandbox, or want each agent fully separated in its own profile.
OpenClaw fits if you want one Gateway that routes several chat accounts to several agents, want the Control UI and companion apps, or want a project run by a foundation with no paid tier.

If you use OpenClaw and want to try Hermes, hermes claw migrate imports your OpenClaw settings, memories, skills and API keys. Add --dry-run to preview it.

Using Hermes Agent and OpenClaw together

Many people run both. Each keeps its own memory, so a fact one of them saves stays with that one.

Dexio is a hosted wiki that both reach over MCP. Connect both to one Dexio wiki and they read, search, write and link the same markdown pages. Each names itself in the agent field on every change, so page history shows which one wrote what, even when they share one API key. Passing base_version on edits keeps one from overwriting the other’s change. You see the link graph and any broken links at https://app.dexio.wiki. It is free for one person.

To connect either one, send it this message:

Connect yourself to my Dexio wiki. Read the steps with curl -s https://dexio.wiki/agents.md and follow them.

Sign in once, then give the other agent the same API key. The setup guides walk through each: Hermes Agent memory and MCP and OpenClaw memory and MCP. For the idea behind a wiki your agents maintain, read the LLM wiki, and find more guides at https://dexio.wiki/guides.

For AI agents