guide

OpenAI dots vs OpenClaw vs Hermes Agent: who runs it?

Dots runs the cloud computer for you. OpenClaw and Hermes Agent give you more control, plus the upkeep. Compare costs, model choice and ChatGPT plan coverage.

EiliyaSeptember 30, 20268 min read

OpenClaw and Hermes Agent both appear in OpenAI's Sign in with ChatGPT directory. OpenClaw is an open-source integration; Hermes Agent's listed route goes through Nous Portal.

That's the first detail to check in an OpenAI dots vs OpenClaw comparison. Running an open-source agent and subscribing to a hosted service are different decisions, even when both can involve your ChatGPT plan.

Dots adds a managed OpenAI agent to eligible ChatGPT accounts. OpenClaw and Hermes let you assemble more yourself, including the model route and running environment. Ask who you want operating that system before comparing demos.

OpenAI dots vs OpenClaw and Hermes at a glance

Question	OpenAI dots	OpenClaw	Hermes Agent
Who builds it?	OpenAI	Open-source OpenClaw project	Nous Research
Where does it run?	OpenAI cloud computer; optional local access	Your computer or a server you operate	Local machine or server, with several execution backends
Model choice	GPT-6 Astra	Configurable providers, including custom endpoints	Multiple providers, including custom compatible endpoints
Agent software cost	First dot included in eligible plans	Open-source software; model and hosting costs remain	MIT-licensed software; model, tool and hosting costs remain
ChatGPT plan route	Native ChatGPT product	Listed as an open-source Sign in with ChatGPT integration	Nous Portal is the listed plan-usage route
Recurring work	Persistent assignments and scheduled tasks	Always-on gateway and connected messaging	Gateway, cron schedules and persistent skills
Main operational burden	Permissions and review	Gateway, credentials, updates and access controls	Runtime, providers, credentials, updates and access controls

The table compares how the systems work. There's no independent test here showing that one handles every browser or messaging task more reliably than the others.

Dots takes the cloud computer off your hands

Dots runs on GPT-6 Astra and has a persistent cloud computer. OpenAI says its plugins connect to more than 4,000 apps, with conversations in ChatGPT, Slack and Teams. It can keep working after you leave, retain notes and return to an ongoing responsibility.

Your first dot is included in eligible plans. Personal access starts at Pro 100, priced at $100 per month, and excludes the EEA, UK and Switzerland. Business Premium and Enterprise follow different availability rules, so your location and workspace may settle the choice before model preferences do.

OpenAI operates the cloud environment and supplies the permission system; you can inspect the computer and take control. Its documented safeguards include read-only proactive research and confirmations for specified actions. Password changes and transfers between financial accounts must go back to you.

That reduces what you need to build, while keeping you inside OpenAI's product boundaries. You can't replace Astra with any provider you like or edit the official agent runtime.

Included access has usage conditions. Dot conversations and Codex Work tasks get different treatment: delegated Codex Work counts toward its usual limits. Don't assume a ChatGPT subscription covers every bill an agent might generate.

OpenClaw gives you the gateway to run

OpenClaw is a personal assistant you run on your own machine. Its website describes handling email, managing calendars and checking in for flights through connected tools. Messaging channels let you reach it from services you already use.

The gateway connects the assistant, model provider and channels. Getting started means installing the software, configuring it with openclaw onboard and opening a dashboard. If you're comfortable operating a service, the setup is manageable; maintenance is still yours.

Want it reachable while your laptop sleeps? You'll need another machine or a server that stays available. A powered-off computer won't process messages, and a server adds hosting expenses and account administration.

OpenClaw documents custom providers with a base URL, API key and API transport. A service exposing compatible Chat Completions can use its openai-completions route. For Responses transport, the backend must actually implement Responses; changing a label won't add support.

You get provider choices that dots doesn't offer and control over where inference happens. You also get the compatibility checks: tool calling, model identifiers and supported endpoints must match your chosen provider.

Hermes puts reusable skills at the center

Hermes Agent is an MIT-licensed project from Nous Research. Its README describes an agent that builds skills from experience, improves them during use and searches past sessions. Those are design claims; you still need to test whether a saved skill is reliable.

Reusable procedures and persistent memory are central to the system. If you repeat a technical workflow and refine it over time, that's a reason to inspect Hermes.

The docs offer desktop and CLI setup routes. Use hermes model to select a model provider and hermes gateway setup to configure messaging for an always-on deployment.

The gateway supports Telegram, Discord, Slack and WhatsApp, among other channels. Built-in cron scheduling can deliver results to a configured platform.

Hermes supports terminal execution backends including local execution, Docker and SSH. Those choices determine where commands and files live. Decide where work executes before handing over credentials; installing the chat interface only covers part of the setup.

Nous Portal is an optional provider route. The open-source agent can also use direct providers or a custom OpenAI-compatible endpoint. Buying a Portal subscription and running Hermes yourself are separate decisions.

A ChatGPT login doesn't pay for every deployment

OpenAI's DevDay recap names OpenClaw among its plan-usage partners, and its directory lists Hermes Agent through Nous Portal. These are supported routes, not blanket coverage for every model request in every deployment.

Eligible Plus and Pro users can grant a supported app access to ChatGPT plan usage. Work and Codex allowances are shared, with per-app weekly caps; each integration doesn't get a fresh pool. The app may still charge a subscription or infrastructure fee.

Follow the integration flow for your route. The directory puts OpenClaw in the open-source section and Hermes through Nous Portal in the apps section. Downloading Hermes doesn't grant the same commercial plan arrangement as signing in to Portal.

Sign-in identity and permission to use plan allowances are separate too. OpenAI says the usage permission doesn't expose your ChatGPT conversations or memory to the app. The agent and its connected services still govern its own chats, files and credentials.

Self-hosting doesn't keep every request local

Self-hosting lets you control the agent's runtime and storage location. Choose a hosted model, though, and information in model requests leaves your machine. Connected apps add their own data routes, so self-hosting alone doesn't keep prompts private from an external inference provider.

OpenClaw defines one trust boundary per gateway, intended for a trusted operator or team. Its security docs warn against treating a single gateway as a secure boundary between hostile users. If unrelated people will use the system, separate credentials and environments matter.

Hermes documents command approval modes, user authorization and isolation options, with cautious defaults for dangerous unattended cron commands. You'll still need to understand which tools and credentials you enabled.

Dots puts more infrastructure in OpenAI's hands. The company says business data isn't used for training by default and personal accounts follow their data controls. That can simplify administration, but OpenAI manages the environment and sets its rules.

Choose by the work you're willing to own

Pick dots if you already qualify, want a managed cloud computer and prefer reviewing permissions to operating a gateway. Start with a bounded assignment before relying on it for a business process.

Pick OpenClaw if you need configurable providers and your own deployment. Budget time for updates, authentication and channel setup alongside model costs.

Pick Hermes if persistent skills, session search and execution choices fit your work. Test whether a saved procedure improves your repeated task; memory alone doesn't establish that.

For either open-source option, test the whole workflow: receive a request, access the right source, ask for approval, produce a result and resume after a restart. A good first chat doesn't verify those later steps.

AI Pass: a model route and earnings for app developers

OpenClaw's custom provider makes AI Pass a possible route via https://aipass.one/v1, an AI Pass key and a supported model. That's an interface match, not an integration test; hosting and tool access remain separate. Independent of OpenAI, AI Pass offers a user-funded, multi-provider, pay-as-you-go wallet.

With Sign in with ChatGPT, the user's plan pays: your app's AI is effectively free to them, and you earn nothing on that usage. AI Pass users pay from their wallet; you can add a markup and earn on eligible paid usage under the current AI Pass terms. Start today at the docs: paste the brief into your coding agent, let it inspect and set up your project, then sign in and approve in your browser.

FAQ

Is OpenClaw a free alternative to dots?

The software is open source. You still fund model usage, hosting and any paid connected services, and take on more operational work.

Can Hermes Agent use my ChatGPT subscription?

OpenAI lists Hermes Agent through Nous Portal among apps with ChatGPT plan usage. Follow that route and its terms; self-hosted configurations don't get blanket coverage.

Does self-hosting keep all agent data local?

Only if your whole configuration keeps it local. Hosted models receive inference requests, and external tools receive the data their actions need.

Which is the best always-on AI agent in 2026?

There's no universal winner in this comparison. Dots suits managed deployment, OpenClaw gives you gateway and provider control, and Hermes focuses on persistent skills and flexible execution.