LATEST
AI Tools

MyClaw 3.0: AI Teammates on Any Harness, From $20 a Month

Updated November 22, 2026
15 min read
Editor's Note: This article was human-reviewed for accuracy
Colleagues working together on laptops in a bright office

Quick Answer & Key Takeaways

  • • MyClaw 3.0, launched November 18, 2026, sells ready-made AI teammates that run on Claude Code, Codex, Hermes, Pi or OpenClaw, with shared memory and a built-in data layer the company says covers more than 5,000 sources.
  • • Plans start at $20 a month for 4,000 credits on a 4 vCPU cloud computer, rising to $200 a month for 40,000 credits on a 16 vCPU machine, with annual billing cutting about 17%.
  • • There is no free trial, and the pricing page publishes no rate for what a credit buys, so the real monthly cost of a working teammate is still an open question.

MyClaw 3.0 arrived on November 18, 2026 with a pitch that is easy to understand: stop building an AI agent from scratch and hire one instead. The launch post from MyClaw, a managed hosting service for the open-source OpenClaw agent, promises ready-made teammates with tools built in, running on your choice of agent harness, on a cloud computer that keeps working after you close your laptop. By Friday morning the post had passed 1.3 million impressions on X, according to independent coverage by i10x, a large reach for an account with roughly 4,000 followers.

The idea lands at a moment when the agent market is splitting into layers. Anthropic sells the model, Cognition sells managed coding agents, and MyClaw is betting the valuable layer is the staffing agency: the teammate with tools, data and a computer, ready to work. This review looks at what 3.0 actually ships, what it costs line by line, and the one question the pricing page does not answer.

What MyClaw 3.0 actually is

MyClaw is a managed hosting service run by Cubo World Inc. It takes open-source agent software, mainly OpenClaw and the Hermes Agent, and runs it for you on a private cloud computer: always on, updated, backed up daily, reachable from chat apps. Its About page describes the service as the easy button for self-hosted AI, and states plainly that MyClaw is not affiliated with, endorsed by, or officially connected to the OpenClaw project or its maintainers. OpenClaw itself is free and open source under the MIT licence, with about 395,000 stars on GitHub as of November 22, 2026. We covered the self-hosted version in our OpenClaw review, including its own README warning to treat inbound messages as untrusted input and set up sandboxing before connecting other people.

Version 3.0 moves the product up a level, from hosting one agent to assembling a team. Instead of a blank agent you configure from scratch, you pick a ready-made role, choose which agent engine runs it, and give it tools and data. The company calls these raised teammates, and the launch post lists four selling points: any harness, more than 5,000 live data sources, a cloud supercomputer, and any channel. The product follows a wider industry push toward persistent, role-based agents, from Cognition Devin Bots, which own goals and lead teams, to Microsoft Teams letting agents join as digital teammates.

One clarification before anything else: MyClaw (myclaw.ai, by Cubo World Inc.) is not MyClawn (myclawn.com). The two names are one letter apart, the companies are unrelated, and neither is a rebrand or a fork of the other. MyClawn gives an agent its own cloud desktop that you watch live; MyClaw hosts agent frameworks such as OpenClaw and Hermes and reaches you through chat channels. If you land on the wrong site, you are reading about the wrong product.

What changed in version 3.0

Three things are new, and one of them is the price list itself. First, the role templates. MyClaw 3.0 ships eight ready-made roles: Chief of Staff, Researcher, PR Reviewer, Project Manager, Content Creator, Design Handoff, Software Engineer and SEO Content Strategist. Each template comes with its own role, runtime, tools and skills, and you can edit it after picking it. Second, the harness choice. A teammate can run on OpenClaw, Hermes Agent, Pi, Claude Code, Codex or a DeepSeek harness, and MyClaw says the teammates share context across harnesses, so work can move between engines without starting from scratch. Third, the data layer. A built-in Data Router gives agents search and SEO, social, company, ad, market and news data, with no API keys to manage and pay-as-you-go billing.

The fourth change is quieter but matters more to your wallet: the old price list is gone. Before 3.0, MyClaw sold four tiers by machine size, Lite at $29 a month, Pro at $59, Max at $119 and Ultra from $239, with AI token usage billed on top. The pricing page now lists three credit plans, Pro, Max and Ultra, plus a custom Enterprise tier, and the Lite tier has disappeared from the page. That is a repricing as well as a relaunch, and anyone comparing old reviews against the current page will find different numbers. The old prices are still visible in reviews from October and early November, so check the date on anything you read.

How the teammates work

MyClaw describes its product in four layers, and the structure is worth understanding because it decides what you are actually buying. The agent is the teammate you configure. Its runtime, OpenClaw, Hermes Agent, Pi, Claude Code, Codex or a DeepSeek harness, powers how it works. Its model provides the intelligence behind its responses, chosen from more than 50 models, with the available choices depending on the runtime you select. The template gives the agent its role. Shared memory is the layer that ties a team together: MyClaw promises one memory across your agents, so teammates can hand off work without starting from scratch.

In practice, a teammate lives on a dedicated cloud computer that browses the web, works with files and runs recurring tasks overnight. You talk to it from the browser, Slack, Teams, Telegram or WhatsApp, and the company says a teammate is ready in about 30 seconds. Agents can draft emails, update projects, research a market, build a website or collect invoices on a schedule, and the site says work continues after you disconnect. What the launch materials do not show is any benchmark, customer result or worked example of a teammate completing a real job. Every performance and data claim so far comes from the company itself.

Template roleWhat it is for
Chief of StaffDaily briefs, meeting prep, follow-ups and inbox triage
ResearcherMarket research, cited briefs and monitoring a topic over time
PR ReviewerReviewing drafts and messaging before they go out
Project ManagerStandup prep, sprint updates and recurring project follow-ups
Content CreatorDrafting posts, campaigns and content from a brief
Design HandoffPreparing design work for developers
Software EngineerCode review, test writing, debugging and pull request prep
SEO Content StrategistAudits, briefs, reports and content refresh work

MyClaw 3.0 pricing, line by line

Every plan includes one always-on private agent with OpenClaw or Hermes, and model usage is charged against the monthly credits. The figures below are from the official pricing page, read on November 22, 2026. Conversions use November 22, 2026 market rates of 1 USD = 0.76 GBP and 1.43 CAD, marked as approximations.

PlanMonthly priceAnnual priceCredits per monthCloud computer
Pro$20 (about £15.28 / C$28.68)$17 a month, $200 a year (about £12.99 / C$24.38)4,0004 vCPU, 8 GB RAM, 80 GB SSD, 5 TB traffic
Max$40 (about £30.56 / C$57.37)$34 a month, $400 a year (about £25.98 / C$48.76)8,0008 vCPU, 16 GB RAM, 160 GB SSD, 10 TB traffic
Ultra$200 (about £152.80 / C$286.84)$167 a month, $2,000 a year (about £127.59 / C$239.51)40,00016 vCPU, 32 GB RAM, 320 GB SSD, 20 TB traffic
EnterpriseCustomCustomCustom64 vCPU, 128 GB RAM, 1.28 TB storage

Annual billing cuts the monthly price by about 17%, and you can upgrade, downgrade or cancel at any time, with price differences prorated for the remaining billing period. Payment is by major credit and debit cards through Stripe, with Apple Pay and Google Pay also supported. There is no free trial. The pricing page FAQ says so directly, and offers a 7-day money-back guarantee instead: a full refund if you are not satisfied within the first week, then cancel anytime with no contracts.

Two credit rules deserve attention. First, the per-credit price is flat. Our arithmetic from the published figures puts every tier at $0.005 per credit: $20 for 4,000, $40 for 8,000 and $200 for 40,000. Paying more buys a bigger machine and more credits, not cheaper credits, so the reason to move up a tier is volume and compute, not a better rate. Second, subscription credits expire at the monthly refresh, while one-time purchased credits carry over. You can also buy extra credits in a one-time pack without changing your plan, and you can bring your own API keys instead, for providers such as OpenRouter, OpenAI, Anthropic and Nous Portal, or any compatible endpoint.

The gap in the price list is what a credit buys. The pricing page publishes no per-model or per-task credit rate, and the site FAQ tells you to check the model usage charges for your setup before you begin. That means the $20 Pro plan has no published answer to the only question that matters for budgeting: how many real tasks does 4,000 credits cover. The Data Router is pay as you go on top, with no published per-call rate either. Until MyClaw publishes a rate card, the honest way to read the price is as a machine rental with a metered model bill attached, and the meter has no visible dial.

MyClaw 3.0 vs the alternatives

The comparison that matters most is the one MyClaw invites: run OpenClaw yourself for free, or pay MyClaw to run it for you. Self-hosting costs nothing for the software, which is MIT licensed, but you pay for your own machine, your own model bill, and your own time keeping it updated and online. MyClaw charges $20 a month and takes the maintenance, the uptime and the backups. That trade is the whole product for a personal assistant. The 3.0 templates and the multi-harness choice are the reasons to pay for more than the cheapest tier, and the credit system is the reason to read the fine print before you do.

OptionWhat you getEntry priceModel bill
Self-hosted OpenClawFree MIT-licensed agent on your own machine; you maintain it$0Your own API keys, billed by the provider
MyClaw 3.0 ProHosted, always-on agent; 4,000 credits; 4 vCPU machine$20 a month ($17 annual)Charged against monthly credits, or your own keys
Claude Code ProAnthropic subscription coding agent; model usage included$20 a month ($17 annual)Included in the subscription
Devin Bots (Cognition)Managed coding agents that own goals and lead teams$200 a month (Max plan, Bots included)Included in the Max plan allowance

Two rows in that table deserve a note. Claude Code Pro costs the same $20 a month as MyClaw Pro, but the money buys different things: a subscription to Anthropic's coding agent with model usage included, against a hosted machine plus a credit allowance that your model usage draws down. MyClaw can also host Claude Code as one of its harnesses, so the two are complements as much as rivals. Devin Bots sits at the other end: $200 a month is the entry point that includes Bots, which is exactly the price of MyClaw Ultra. One buys a team of coding agents with a track record of shipped pull requests; the other buys a bigger machine and 40,000 credits for agents you configure yourself. We covered both in our Claude Code explainer and our Devin Bots review.

The catches

The credit question has no published answer

This is the central catch, and it is the one MyClaw's own pages confirm by omission. No per-model credit rate, no worked example of a task, no table showing what 4,000 credits covers. The company says lower-cost models reduce credit usage, and that model choices depend on the runtime, but a buyer cannot turn that into a monthly budget before paying. i10x reached the same conclusion in its November 20 analysis: how far 4,000 credits go in real work is not spelled out on the pricing page. Treat the first month as a paid experiment, and watch the usage dashboard before committing to annual billing.

No independent proof yet

There are no benchmarks, no customer results and no independent hands-on review of the teammates at launch. The 5,000-plus data sources figure is not broken down publicly, the shared-memory claim has no published detail on how context actually transfers between very different harnesses, and the launch post gives no customer numbers. The X conversation around the launch included near-identical promotional replies from small accounts, which i10x flagged as a reason to wait for independent reviews before judging quality. Elvis Saravia of DAIR.AI highlighted the harness flexibility as the real idea, writing that agents should not be locked into a single harness. That is a fair reading of the pitch. Whether the teammates do reliable work is still unproven.

Two small inconsistencies on the site itself

The pricing page call to action reads Start Free Trial, while the same page FAQ says the company does not offer a free trial at the moment. The FAQ also still mentions the Lite tier in several answers, for example calling Lite enough for many personal-use workflows, even though Lite no longer appears on the pricing page. Neither issue changes what the product does, but both suggest the site was updated in a hurry for the 3.0 launch, and a buyer should trust the pricing page figures over the FAQ when the two disagree.

The security question is yours, not theirs

MyClaw says each instance is isolated, that it does not access, share or train on your data, and that your memory, conversations, skills and API keys are separated from other customers. Those are the company's own claims, and there is no published independent security review. The agents sign in to your tools and act in your accounts, so the permissions you grant decide the risk. OpenClaw's own README warns to treat inbound messages as untrusted input and set up sandboxing before connecting other people, and that warning applies with more force, not less, when the agent runs on someone else's cloud and talks to your Slack and WhatsApp. Start with read-only routines and human-approved drafts, and keep payment access and account changes behind approval until you have seen the agent work.

Who should buy it, and who should skip it

MyClaw 3.0 fits a specific buyer: someone who wants an OpenClaw or Hermes agent running around the clock, does not want to maintain a server, and values the ready-made roles more than the lowest possible price. A solo founder who wants a researcher that monitors a topic overnight, a small team that wants a Chief of Staff in Slack, or a developer who wants Claude Code available without keeping a laptop awake are the natural customers. The Pro plan at $20 a month is the sensible starting point, since the company's own guidance calls Pro the best default for coding and customer support workflows.

Skip it for now if you need a predictable monthly cost, if you want proof the teammates do reliable work, or if you already run OpenClaw happily on a machine you own. The credit system makes the first month a measurement exercise, and there is no free trial to run it in. If your main need is a coding agent rather than a general teammate, a Claude Code Pro subscription at the same $20 a month includes the model usage, which is a cleaner deal for that one job.

Common mistakes

First, do not buy the annual plan on day one. The 17% saving is real, but the credit burn rate is unknown until you have run real work for a month, and the 7-day money-back window is the only risk-free period. Second, do not connect the agent to accounts it can spend from until you have tested it on read-only work. Drafts and summaries first, sends and purchases after you have seen the output. Third, do not assume the template does the setup for you. A Researcher template still needs your sources, your criteria and your review rules before its output is worth trusting. Fourth, do not pick a harness for the brand name. The runtime decides what the agent can do, and the model decides how well it does it, so match the harness to the job: Claude Code or Codex for code, OpenClaw or Hermes for personal routines. Fifth, check whether your own API keys are cheaper than credits for your model of choice. MyClaw lets you bring keys, and for heavy use on a cheap model, the provider's own bill may undercut the credit meter.

FAQ

What is MyClaw 3.0?

MyClaw 3.0 is a November 18, 2026 release from Cubo World Inc. that turns its managed OpenClaw hosting service into ready-made AI teammates. You pick a role such as Researcher or Chief of Staff, choose an agent harness such as Claude Code, Codex, Hermes, Pi or OpenClaw, and the agent runs on a private cloud computer with shared memory and a built-in data layer.

How much does MyClaw 3.0 cost?

Pro costs $20 a month, or $17 a month billed annually at $200 a year, and includes 4,000 credits a month on a 4 vCPU machine. Max costs $40 a month, or $34 annual, with 8,000 credits on 8 vCPUs. Ultra costs $200 a month, or $167 annual, with 40,000 credits on 16 vCPUs. Enterprise pricing is custom. Annual billing saves about 17%.

What does a MyClaw credit buy?

MyClaw has not published a per-model or per-task credit rate. Model usage is charged against your monthly credits, lower-cost models reduce credit usage, and one-time purchased credits carry over while subscription credits expire at the monthly refresh. You can also bring your own API keys for providers such as OpenRouter, OpenAI, Anthropic and Nous Portal instead of using credits.

Is there a free trial?

No. The pricing page FAQ says MyClaw does not offer a free trial at the moment, and points to a 7-day money-back guarantee instead. Payment is by major cards through Stripe, with Apple Pay and Google Pay supported. Note that the page call to action still reads Start Free Trial, which contradicts the FAQ.

Is MyClaw the same as OpenClaw?

No. OpenClaw is a free, open-source agent under the MIT licence, with about 395,000 GitHub stars. MyClaw is an independent paid service by Cubo World Inc. that hosts OpenClaw and other agent runtimes on its own cloud computers. MyClaw states it is not affiliated with, endorsed by, or officially connected to the OpenClaw project or its maintainers.

Can MyClaw agents work in Slack and WhatsApp?

Yes. The launch announcement lists the browser, Slack, Teams, Telegram and WhatsApp as channels, and the site FAQ lists the web, Telegram, WhatsApp and Slack. Agents can also use your tools and files, depending on the connections and permissions you configure, and they keep working after you close your laptop.

MyClaw 3.0 is a sensible answer to a real annoyance: setting up an agent with the right tools, data and a machine that stays on is still fiddly, and a pick-a-role, pick-an-engine product removes most of that friction. The $20 entry price is fair for a hosted, always-on machine. What is missing is proof: proof the teammates do reliable work, and a published rate that turns credits into a budget. Until both appear, buy one month, measure the burn, and let the usage dashboard decide whether the teammate earns its keep.

Sources

  • MyClaw official homepage: https://myclaw.ai/
  • MyClaw official pricing page: https://myclaw.ai/pricing
  • MyClaw official About page: https://myclaw.ai/about
  • i10x, MyClaw 3.0: AI Teammates, Features and Pricing (November 20, 2026): https://i10x.ai/news/myclaw-3-0-ai-teammates-claude-code-codex-hermes-openclaw
  • OpenClaw Easy, MyClaw Review 2026 (prices checked October 15, 2026): https://openclaw-easy.com/blog/myclaw-review-2026.html
  • GitHub, OpenClaw repository (stars and forks verified November 22, 2026): https://github.com/openclaw/openclaw
  • Anthropic, Claude pricing page (verified November 22, 2026): https://claude.com/pricing
  • MyClawn, MyClaw and MyClawn are different products: https://www.myclawn.com/myclaw-vs-myclawn
  • AI Core Updates, OpenClaw: The Free AI Assistant You Text From WhatsApp: https://aicoreupdates.com/article/openclaw-the-free-ai-assistant-you-text-from-whatsapp
  • AI Core Updates, Devin Bots: The AI Manager That Runs a Team of Coding Agents: https://aicoreupdates.com/article/devin-bots-the-ai-manager-that-runs-a-team-of-coding-agents
  • AI Core Updates, Claude Code: Anthropic's Agentic Coding Tool, Explained: https://aicoreupdates.com/article/claude-code-anthropic-s-agentic-coding-tool-explained
15 min read
5 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more