Claude Duet for Small Business (2026): Pricing, Limits, Permissions, and Workflow Fit

Claude Duet is worth joining the waitlist for if a US-based owner already pays for Claude Pro or Max and wants to test one narrow, read-heavy workflow. It is not ready to become a small business's unattended operating layer. During the research preview, Duet has no separate add-on price, but it shares the plan's five-hour session and weekly usage limits. Morning briefs, periodic checks, and background work all consume that allowance. Access is waitlist-gated, US-only, and unavailable on Team and Enterprise plans.

That makes the real Claude Duet pricing question less about a new subscription and more about capacity, permissions, and reliability. Pro costs $20 monthly or $200 annually; Max starts at $100 monthly, with 5x and 20x usage options. Duet can work across connected apps, maintain one persistent thread, use its own cloud computer and browser, prepare files and tasks, and request approval for consequential actions. It also remains an early preview that checks at intervals, cannot show or hand over its browser, and can misread context.

This is a vendor-document-verified buyer guide, not a hands-on test. Product behavior, access, pricing posture, usage limits, connectors, approvals, and comparison-product boundaries were checked against current vendor documentation on October 25, 2026. No Google Search Console or GA4 evidence was available, and no performance claim is inferred.

Duet's practical value is not that it can touch many apps. It is that it can prepare work across them while the owner retains control of consequential actions.

Claude Duet availability and pricing: facts checked October 25, 2026

Duet is included at no extra charge with eligible personal Claude plans, but there is no fixed Duet task quota to budget around. Anthropic says it uses the same plan limits as ordinary Claude activity, including five-hour session and weekly limits.

Anthropic does not publish a universal number of Duet briefs, checks, or completed tasks per plan. A short inbox scan and a financial model do not consume the same amount of work. Larger assignments use more model turns and background activity, while more frequent monitoring spends allowance even when nothing urgent happens.

Do not buy Max solely because “20x” sounds safer. First measure whether Duet improves one workflow enough to justify the opportunity cost of using the same allowance you need for Claude chat, Cowork, or Claude Code.

What does Claude Duet do differently from ordinary Claude chat?

Ordinary Claude chat waits for a prompt; Duet proactively checks connected work, creates a morning brief, and returns when a task finishes or needs a decision. It uses memory, skills, and connectors from the same Claude account, but the working model is different.

• One persistent relationship: text or voice messages, quick requests, and longer projects can stay in one ongoing thread. Separate channels can organize major topics.
• Proactive briefings: Duet prepares a morning summary and checks connected accounts during the day.
• Interval monitoring: checks happen every few hours or at a cadence you request, not continuously or in real time.
• Background execution: its cloud computer keeps working when the owner's laptop or phone is off.
• Browser work: it can use a cloud browser on sites without connectors, but the user cannot watch or take over that browser during the preview.
• Task and file outputs: a Tasks tab separates running work from the main thread, and Duet can return documents, sheets, decks, dashboards, and other files for review.
• Approval gates: Duet is designed to stop before sending messages, making payments, signing documents, or deleting anything.

The Claude Marketplace has more than 6,000 connectors and plugins. That number describes the ecosystem, not 6,000 equally mature Duet workflows. Connector behavior can be narrower inside Duet—Anthropic specifically warns that Slack has limitations—and each connected account still has its own scopes and source-system permissions.

Which small-business apps and workflows fit Duet?

Duet fits cross-app preparation and exception spotting better than irreversible transaction processing. The strongest early workflows produce a reviewable draft, list, or brief before anything leaves the business.

Duet also supports purchases through Link, Stripe's wallet, and Shop Pay. Anthropic says it asks before each purchase. Link presents another approval and uses a one-time virtual card so the merchant does not receive the real card number. That is safer than silently storing a card in an agent, but it is not a reason to remove a business spending policy.

Microsoft 365 is the important exception. Although Microsoft 365 has appeared among product-page integration branding, Anthropic's current Duet Help Center explicitly says Microsoft 365 is not available yet. The ordinary Claude Microsoft 365 connector can be available elsewhere in Claude; that does not make it a working Duet connector today.

What permissions should a small business give Claude Duet?

Start with read and prepare permissions, require explicit approval for every external or financial action, and prohibit anything that cannot be safely reversed. The product's approval classifier is a safeguard, not your complete control system.

A practical Duet policy has four zones: observe, prepare, approve, and prohibit.

Also prohibit access to someone else's account, broad administrator credentials, credential vault exports, tax submissions, and unreviewed legal commitments. Those controls are operating policy recommendations, not claims that Duet supports each action.

How should you budget Claude Duet usage?

Budget by workflow intensity and observed allowance consumption, not invented tasks-per-month estimates. Duet shares the same limit pool with Claude, and background checks can consume it before the owner starts a normal chat.

Record four things each day: plan usage before the run, plan usage after, accepted outputs, and operator corrections. If a Max tier only masks a bloated workflow, the business has purchased a larger bucket for a leaking pipe.

Claude Duet vs cloud Cowork: which should you use?

Choose Duet for a proactive, persistent assistant that brings issues to you; choose cloud Cowork for deliberate tasks you start, steer, and review as sessions. Both can run on Anthropic's infrastructure and use connectors, files, skills, browser capabilities, and notifications, but they are not interchangeable.

Ordinary Claude chat is still the simplest option for a one-off answer, analysis, or draft. Dispatch is a Cowork agent that can split an outcome into child Cowork or Code tasks; its desktop-hosted phone workflow needs the computer awake, though cloud Cowork is now the simpler route for many users. Remote Control is specifically a window into a Claude Code session running on your own machine, so the machine and Claude process must stay on.

The labels are close enough to make a family reunion awkward. The operating boundaries are what matter.

How does Duet compare with OpenAI Dots and Meta Muse?

Dots and Muse are relevant alternatives, not feature-for-feature substitutes. Compare the current product, plan, region, connectors, permissions, and business data policy you can actually obtain—not the most ambitious launch-page sentence.

• OpenAI Dots: OpenAI describes Dots as always-on agents with a cloud computer, browser, more than 4,000 app plugins, proactive read-only research, Custom Rules, Activity View, messaging in ChatGPT, Slack, and Teams, and voice calls. Dots are rolling out across Pro, Business Premium, and Enterprise in eligible markets. The first dot is included with eligible plans, and its conversations do not consume ordinary ChatGPT usage limits, although delegated Codex or ChatGPT Work tasks do. That usage model is materially different from Duet's shared Claude limits.
• Meta Muse: Meta describes Muse as a proactive personal agent running in a dedicated secure VM, available in the US and Canada, with messaging through its app or WhatsApp. Muse can use a browser, keep working in the background, ask before sends and purchases, and connect business tools plus Meta surfaces. Meta says most use is free, with subscriptions for more. Muse has a stronger Meta/consumer ecosystem angle; its connector set, subscription limits, and business controls should be verified for the exact workflow.

Neither comparison proves that Duet, Dots, or Muse is “better.” Duet may fit a business already standardized on Claude connectors and memory. Dots may fit a ChatGPT-centered operation that needs Slack, Teams, or voice-call access. Muse may fit a Meta-heavy storefront or creator business. All three are young agent products whose current limits deserve more weight than their roadmaps.

What can go wrong with Claude Duet?

The main risks are missed timing, incorrect judgment, over-broad data access, prompt injection, and preview instability. Approval prompts reduce harm only when the reviewer understands what is being approved.

• Not real-time: an urgent email or payment exception can arrive between interval checks. Duet should not replace alarms, fraud systems, or on-call monitoring.
• Preview mistakes: Anthropic says Duet may misread a thread or misjudge urgency. Expect false positives, missed context, and drafts that need correction.
• Prompt injection: email, web pages, and connected content can contain malicious instructions. Anthropic says no defense is perfect.
• Memory after disconnection: disconnecting an app stops future access, but Duet can retain memory of information used in earlier conversations. Deleting a connection is not the same as erasing all remembered context.
• No browser view or takeover: the owner cannot inspect the browser live or step in to fix a delicate flow.
• No calls or outbound texts: Duet can accept voice messages, but it does not call or text other people. Approved email is the documented outbound path.
• No Microsoft 365 yet: do not plan an Outlook, Teams, OneDrive, or SharePoint workflow around Duet until Anthropic documents support.
• No model choice: Opus 5.5 and automatic effort are fixed during the preview.
• Prohibited finance actions: securities trading and transfers between accounts remain off-limits even with approval.

Use source citations in Duet outputs when available. Anthropic says summaries and numbers identify the email, file, or spreadsheet they came from. The reviewer should open that source before approving a payment, payroll, contract, refund decision, or external message.

Buy, wait, or skip Claude Duet?

Join the waitlist if the marginal subscription cost is zero and you have a safe pilot; wait if access, governance, or Microsoft 365 matters; skip if the workflow needs deterministic real-time automation.

Join the waitlist and pilot now if

• the owner already has Claude Pro or Max in the US;
• one useful workflow can stay read-only or draft-only;
• the business can tolerate preview errors and interval checks;
• one accountable reviewer can inspect every external or financial action; and
• usage can be measured against existing Claude work.

Wait if

• the business uses Team or Enterprise and cannot mix a personal-plan agent into company operations;
• Microsoft 365 is the core system;
• a visible browser takeover is required;
• the business needs calls, outbound texts, or real-time monitoring; or
• policy requires formal admin controls, audit coverage, retention controls, or centralized access management not yet available in Duet.

Skip Duet for this workflow if

• the process must execute at an exact time with guaranteed sequencing;
• a missed event creates safety, legal, payroll, or financial harm;
• the job involves securities trades or bank-to-bank transfers;
• the workflow requires high-volume record synchronization, idempotent retries, or detailed machine logs; or
• nobody can reliably review approval requests.

Use deterministic automation for exact transaction pipelines. Use Duet for interpretation, preparation, prioritization, and exception handling around those systems.

A low-risk 14-day Claude Duet pilot

A useful pilot proves that Duet can improve one workflow without creating hidden usage or authority. Do not connect ten apps and call the resulting fog “learning.”

The pilot expands authority only after evidence; every stage retains a route back to the manual process.

Days 1–3: define the boundary

1. Choose one owner and one workflow, such as a daily Gmail-and-Calendar brief.
2. Record the manual baseline: preparation time, missed items, corrections, and current Claude usage.
3. List allowed sources, prohibited data, approval owner, and rollback trigger.
4. Connect the narrowest account or source scope available.

Days 4–7: run read-only

1. Request one morning brief at a fixed cadence.
2. Compare every flagged item with the source.
3. Seed a harmless test message containing misleading instructions and confirm Duet does not follow them.
4. Track false positives, missed items, citations, latency, and allowance use.

Days 8–10: add drafts, not actions

1. Let Duet draft replies, invoice reminders, meeting options, or CRM follow-ups.
2. Keep sends, record edits, payments, signatures, payroll, publishing, and deletion manual.
3. Review sensitive-data handling, recipients, totals, dates, and attachments.
4. Correct Duet in the thread and note whether the next output improves.

Days 11–14: test repeatability and decide

1. Run normal, missing-data, duplicate, stale, and permission-error cases.
2. Measure accepted outputs, correction rate, missed-item rate, approval volume, and usage impact.
3. Disconnect the app once and confirm future access stops; separately review and delete remembered context or chats where appropriate.
4. Choose one outcome: stop, narrow, repeat the pilot, or expand to one adjacent draft-only workflow.

Success should mean zero unauthorized actions, zero sensitive-data exposure, no critical missed event, and a meaningful reduction in reviewable preparation work. A polished draft that consumes too much allowance or creates more checking than it saves is not success.

Claude Duet rollback checklist

Rollback must restore the manual workflow, not merely disconnect one connector.

• Pause Duet schedules, briefings, and recurring checks.
• Finish or cancel active tasks and record their status.
• Disconnect each app in Claude settings and revoke access in the source system where available.
• Remove Link and Shop Pay connections and review recent transaction history.
• Rotate any credentials entered into sites if exposure is suspected.
• Review sent messages, signed documents, payments, payroll, published changes, and deletions.
• Restore affected records from trash, version history, backups, or the system of record.
• Review and delete relevant Duet chats or remembered items; disconnection alone does not erase past memory.
• Preserve an incident note with timestamps, sources, approvals, outcomes, and corrections.
• Resume the documented manual process and tell affected staff who owns the queue.

Methodology and evidence limits

This guide is an independent synthesis of current vendor documentation, not a hands-on product review. We checked Anthropic's Duet announcement, product page, Help Center guidance, pricing page, usage-limit documentation, Cowork and Dispatch documentation, and Claude Code Remote Control documentation. OpenAI and Meta first-party product pages were used only to describe Dots and Muse at a high level.

No Duet account was provided, no waitlist invitation was accepted, and no connector, purchase, payroll, browser, or approval flow was executed. Connector names and vendor examples confirm intended product scope, not successful operation in every account. Preview features, eligibility, pricing, limits, and integrations can change quickly; recheck the official pages before purchase or rollout.

FAQ

How much does Claude Duet cost?

Duet has no separate add-on price during the research preview. It is included with eligible Claude Pro and Max subscriptions. Pro is $20 monthly or $200 annually; Max starts at $100 monthly. Taxes and later plan changes can affect checkout pricing.

Does Claude Duet have its own usage limit?

No separate Duet allowance is documented. Duet shares the plan's five-hour session and weekly limits with the rest of Claude. Morning briefs, periodic checks, and background tasks all count.

Can Claude Team or Enterprise accounts use Duet?

Not yet. The research preview is for waitlisted US users on personal Pro and Max plans. Anthropic says it plans to bring Duet to Team and Enterprise later.

Is Claude Duet available outside the United States?

No. Anthropic currently limits the research preview to the US.

Does Claude Duet work with Microsoft 365?

Not during the current preview. Anthropic's Duet Help Center explicitly says Microsoft 365 is not available yet, even though Microsoft branding may appear in broader product or connector marketing.

Can Claude Duet send email without approval?

Anthropic says Duet is designed to show the user what it is about to do and wait for approval before sending a message. Small businesses should verify the exact recipient, content, attachment, and sensitive information every time.

Can Claude Duet issue refunds or run payroll?

Anthropic documents Gusto payroll preparation and execution as an example, but payroll should remain an owner-approved payment workflow. Public Duet documentation does not clearly establish refund execution support, so use it to gather evidence and draft the decision while keeping the refund manual.

Is Claude Duet real-time?

No. It checks connected accounts at set intervals to control usage. An urgent event can arrive between checks.

What happens when I disconnect an app?

Future access stops, but Anthropic says memory of information from earlier conversations is retained. Review remembered context and relevant chats separately from the connector revocation.

Is Claude Duet safer than ordinary browser automation?

It adds cloud isolation, credential handling, approval checks, and prompt-injection defenses, but Anthropic explicitly says no defense is perfect. Consequential work still needs source verification and human approval.