OpenClaw alternatives
OpenClaw alternatives change who runs the agent and who bills its work. As of September 30, 2026, compare Hermes Agent if you want another self-hosted runtime with provider choice. Grok Bot, OpenAI Dots, Claude tasks, and Manus offer managed paths with different accounts, allowances, and access limits. We have not measured a performance winner or compared task invoices.
An agent product supplies the runtime, tools, memory, and permissions around a model. A model API supplies inference. Replacing the model inside OpenClaw is a different decision from replacing OpenClaw itself.
By Glevd. This review covers public documentation and logged-out pricing pages checked .
Choose the part you want to change
Start with the work you want to stop doing. Maintaining a host, managing provider accounts, and approving connected-app actions are different obligations. Our choices below follow documented operating arrangements; they do not establish which agent completes your work most reliably.
- Keep your host and provider accounts: compare Hermes Agent with OpenClaw. Both document provider choice; switching still leaves a runtime to maintain. OpenClaw vs Hermes covers that narrower decision.
- Keep Hermes and consolidate model or tool charges: inspect Nous Portal. Add Hermes Cloud only if you also want a managed host. The Portal subscription and the Cloud meter pay for different things.
- Use an account you already pay for: check Grok Bot, Dots, or Claude access before buying another plan. A qualifying subscription can include agent work, while rollout, geography, and usage limits still decide what you can run.
- Move to Manus for its task workflow: compare the selected credit amount and required add-ons with your tasks. Instinct belongs in a personal-use check only; its public terms exclude the team use case.
Who hosts the agent and chooses the model?
OpenClaw is the baseline in this table. The three Hermes rows separate software, a billing gateway, and hosting. A documented feature describes the provider's stated behavior. Unknown means the inspected sources did not establish it; it does not mean the feature is absent.
| Product / reviewed version | Runtime and host | Model choice and authentication | Access or trust boundary |
|---|---|---|---|
| OpenClaw Baseline: v2026.9.7, released September 30, 2026 OpenClaw release | You run the Gateway on your hardware or a server you choose. Documented OpenClaw FAQ | Choose a provider plugin, API key, or supported subscription login. Local and custom model endpoints are documented; the auth route matters. Documented Provider pluginsCustom providers | A shared Gateway is one trust domain. Configured remote model providers and messaging platforms receive their respective traffic. Documented OpenClaw FAQ |
| Hermes Agent: self-hosted v0.21.5 / v2026.9.24, released September 24, 2026 Hermes release | Run the agent on your own computer or host. The agent software is MIT licensed. Documented Hermes pricing FAQ | Bring provider credentials or use a local endpoint such as Ollama, vLLM, or LM Studio. Model switching works among configured providers. Documented Hermes provider routes | A Nous subscription is optional. Provider accounts, tool services, and your chosen execution environment remain separate decisions. Documented Hermes pricing FAQHermes provider routes |
| Hermes Agent + Nous Portal Same agent; optional model and tool billing route | Portal supplies model and tool access. Buying Portal credits does not by itself move your agent into Hermes Cloud. | Nous OAuth authenticates a selectable model catalog and the Tool Gateway. Nous chooses the backend routing for each catalog model. Documented Nous Portal integration | Selected tool calls use Nous-managed services. You can enable selected gateway tools and retain other configured backends. Documented Nous Portal integration |
| Hermes Cloud Managed hosting path checked September 30, 2026 | Nous hosts a running Hermes instance. Hosting, model inference, and tools have distinct charges. Documented Hermes Cloud billing | Cloud setup lets you choose a model and instance size. Bring-your-own provider keys in that managed deployment were not confirmed. Partly documented Hermes Cloud billing | Deploy through a Nous account with an active subscription or at least $2 in credits. That balance requirement is not a monthly hosting price. Documented Hermes Cloud billing |
| Grok Bot Managed service; controls updated September 28, 2026 Grok Bot controls | The computer runs in Cursor cloud infrastructure. Customer-hosted deployment and custom machine images are unavailable. Documented Grok Bot controls | Cursor selects the served model; there is no customer model picker. Usage analytics identify served models. BYOK was not documented in this check. Partly documented Grok Bot controls | Bots on one account share a computer, browser sessions, files, and logins. Separate Bots do not establish separate security boundaries. Documented Grok Bot FAQ |
| OpenAI Dots Announced September 29, 2026; gradual rollout Dots announcement | Each dot has an OpenAI cloud computer. Access to your local computer is optional and starts off. | The launch names GPT-6 Astra. A customer model picker and BYOK route were not established by the launch or help sources. Partly documented Dots announcementDots rollout and regions | You choose connected apps and permissions and can set rules for actions. Local access requires the ChatGPT desktop app. |
| Claude tasks / former Cowork Unified Claude rollout checked September 30, 2026 | Cloud tasks can continue after you close your laptop. Tasks using local files or apps need Claude Desktop to stay open. Documented Claude task rollout | The Claude model picker remains available. A third-party model or BYOK route for these consumer tasks was not documented. Partly documented Claude task rollout | Task availability is rolling out gradually to Pro and Max accounts. Two accounts on the same plan can have different access. Documented Claude task rollout |
| Manus Manus 2.0 announced September 28, 2026 Manus 2.0 announcement | The 2.0 release includes a managed Cloud Computer option and authorized local computer access. Cloud Computer is a paid add-on. Documented Manus 2.0 announcement | Current customer model selection and BYOK support were not established for Manus 2.0. Earlier release descriptions do not settle either question. Unknown Manus 2.0 announcement | The release describes local execution inside a workspace you authorize. This review did not inspect an authenticated account or its permission settings. Partly documented Manus 2.0 announcement |
| Instinct: personal use Terms revised August 26, 2026 Instinct terms, sections 4 and 6 | The product connects apps and devices. Its public pages did not establish the hosting arrangement or a customer-hosted option. Unknown Instinct product | Model choice, BYOK, and runtime authentication details were not established by the inspected public pages. | The terms restrict the service to personal use. Keep it outside a team or agency shortlist unless the provider supplies different applicable terms. Documented Instinct terms, sections 4 and 6 |
Which account gets the bill?
A subscription price, token rate, daily host charge, and credit allowance use different units. This ledger keeps them separate. It gives you the components to inspect; it does not convert them into cost per completed task. OpenClaw pricing covers the baseline's provider and hosting bill in more detail.
| Product | Billing account and fee | Included usage or service | Additional charges and limits |
|---|---|---|---|
| OpenClaw | No Foundation subscription or Foundation-hosted plan. Model providers, tool vendors, and your host bill their own services. Documented OpenClaw FAQOpenClaw cost accounting | Installing the agent does not establish a hosted model or tool allowance. Any allowance belongs to the configured provider account. Documented OpenClaw cost accounting | Remote inference, paid tools, and hosting depend on your setup. Local usage estimates are not a provider invoice. Documented OpenClaw cost accounting |
| Hermes Agent: self-hosted | No software license charge; model and optional service charges are separate. A Nous plan is not required. Documented Hermes pricing FAQ | Model and tool usage depend on your chosen accounts. The MIT software license is not an inference allowance. Documented Hermes pricing FAQHermes provider routes | Your provider, tools, and hardware or host determine the bill. A local endpoint still uses your own computing resources. Documented Hermes pricing FAQHermes provider routes |
| Hermes Agent + Nous Portal | Portal lists Plus $20, Super $100, and Ultra $200 per month. The inspected plan page displays $ without an ISO currency code. Documented Portal plans | Monthly credits: Plus $22, Super $110, Ultra $220. Rollover caps: $10, $50, $100 respectively. Free has free models and $0 monthly credits. Documented Portal plans | Models and gateway tools spend the Nous balance. Hosting is a separate charge if you choose Hermes Cloud. |
| Hermes Cloud | Hosting spends Nous credits: Medium $0.56/day running; Large $1.09/day running; stopped storage $0.03/day. ISO currency is unconfirmed. Documented Hermes Cloud billing | Running rates cover compute and storage. The service bills daily in UTC, in arrears, using actual running seconds and retained storage. Documented Hermes Cloud billing | Model and tool usage is additional. Stopping the instance retains its workspace and leaves a storage charge. Documented Hermes Cloud billing |
| Grok Bot | Included with qualifying paid Cursor plans; no separate Bot subscription. Eligible individual SuperGrok accounts can be linked. Existing subscriptions retain their own bills. Documented Cursor access and billing | Weekly usage is included. For linked Cursor/SuperGrok subscriptions, the FAQ says the larger grant applies. Grants do not combine; a universal numeric allowance is not published. Documented Grok Bot FAQCursor access and billing | Eligible on-demand usage is charged through Cursor using served-model and token costs. An active task can finish above a configured spend cap. Documented Cursor access and billing |
| OpenAI Dots | First dot included with eligible Pro or Business Premium accounts. Pro is USD $100/$200/$500 monthly. Business Premium lists $100/user/month billed annually or $125 monthly; Enterprise pricing is custom. | Dot conversations do not consume ChatGPT usage. Deeper work has an allowance, extended during the first launch month; no fixed universal task count is established. Documented Dots announcement | Tasks delegated to Codex or ChatGPT Work consume those products' normal usage. Included access does not mean unlimited background work. Documented Dots announcement |
| Claude tasks / former Cowork | Pro is USD $20/month on the US web checkout. Max lists $100/month for 5x or $200/month for 20x; regional and app-store prices can differ. Documented Claude Pro termsClaude Max terms | Conversations and tasks use the same plan allowance. Long tasks can consume more usage; Pro has both five-hour and weekly limits. Documented Claude task rolloutClaude Pro terms | Claude API and Console charges are separate from a paid Claude subscription. This does not turn every Claude integration into the same billing route. Documented Claude API billing |
| Manus | The monthly USD view shows $20/$40/$200 at selected 4,000/8,000/40,000 credits. Annual billing displays $17/$34/$167 per month, billed yearly; the exact annual invoice was not established. | Those selected amounts are monthly credits. Each inspected card also shows 300 refresh credits every day. Daily refresh credits are a separate allowance. Documented Manus pricing selectors | Cloud Computer is a paid add-on; its amount was not established. Customizable credit selections change the displayed subscription price. Partly documented Manus 2.0 announcementManus pricing selectors |
| Instinct: personal use | The terms specify USD for fees, but the inspected pages did not establish a current subscription amount or billing interval. | Included usage, a free allowance, and an overage schedule were not established. Missing published prices do not establish a free service. | Connected-service charges and a complete invoice path were not established from the public pages inspected. |
Two subscriptions can leave two bills without adding the usage grants together. The Grok Bot FAQ, updated September 29, says the larger eligible grant applies; the Cursor plan guide says the grants do not combine and both subscriptions keep billing separately. The numeric grant on a specific account was not measured here. Manus' annual selector also needs care: rounded monthly equivalents do not establish an exact yearly invoice.
Check access before choosing a plan
Dots: Pro rollout excludes the EEA, Switzerland, and the UK. Business Premium follows supported ChatGPT regions. Enterprise beta access starts off and needs administrator enablement. Rollout is gradual, and creation requires the ChatGPT desktop app or desktop web. Business needs at least two paid seats, which can mix seat types.
Grok Bot: paid Cursor individual and self-serve Teams access is documented. Individual SuperGrok, SuperGrok Plus, and SuperGrok Heavy can link; Lite, Team, and Enterprise SuperGrok plans cannot. X Premium+ can also grant usage through a linked account. Cursor Enterprise access goes through the account team.
Claude: the Cowork and chat transition is gradual. Confirm that tasks are available in your account before treating a plan upgrade as access to the unified experience. Local-file work also depends on an open desktop session.
Instinct and Cue: Instinct's public pages leave pricing and account availability unresolved. Manus introduced Cue as an invite-based early-access personal agent. Its ongoing price, model controls, and regional access were not established, so it is a lead to check rather than another complete row here.
How to read and recheck this ledger
We checked first-party documentation, release notes, pricing, and terms on September 30, 2026. The source links sit beside the claims they support. Named release and revision dates describe the source or product version; the check date records this review. Undated pages do not acquire a publication date from being checked.
Manus pricing was inspected in its logged-out browser selectors because the text-only page did not expose the current cards. We checked both monthly and annual views and the selected credit amounts. We did not purchase a plan, connect a personal account, execute a task suite, or reconcile invoices for any product. Performance, reliability, and cost per successful task remain unmeasured.
Before paying, recheck the exact plan, currency, billing interval, allowance, region, and auth route in the linked source and your own checkout. A plan label alone is too little information. For an existing self-hosted setup, changing the model may solve the problem without migrating the runtime: use models for agent tasks, agent benchmark evidence, model API pricing, and the token cost calculator for that decision.
Questions
Is a model API an alternative to OpenClaw?
A model API replaces the inference provider, not the surrounding agent product. OpenClaw still needs a runtime, tools, memory, and permissions. Its provider plugins let you change inference without switching products. Start there if your concern is model behavior or API cost and the existing runtime otherwise suits your work.
Does self-hosting keep all agent data local?
Self-hosting establishes where the runtime runs; it does not establish where every request goes. OpenClaw's FAQ distinguishes local session storage from traffic sent to configured model providers and chat platforms. Inspect the model endpoint, remote tools, and messaging connections before relying on a local deployment to keep a particular workflow on your machine.
Can I compare these subscriptions by the number of credits?
Credit counts are meaningful inside their own billing system. Nous monthly balances and Manus monthly and daily allowances do not establish equal amounts of work. Compare the same tasks, required tools, retries, and final invoices before judging task cost. This ledger records published units; it does not convert unlike credits into a shared performance measure.