Duet vs win.sh: Two Operating Models
Duet lets consultants build an agent once and share it with every client. win.sh runs a 24/7 business loop inside an authority matrix.
Founder, Task Machine
Consultants sell a method: the checklists, templates, and judgment they bring to each client. The question many now face is whether to package that method as an agent their clients can use directly. A founder running their own business faces a different question: whether to let a loop watch the company and act. Duet answers the first and win.sh the second.
Duet lets you build an agent once and share it with every client. You teach it your checklists, templates, and rules, each client gets a workspace of their own with your agent inside, and improvements you make reach every client's copy. Each workspace has an always-on machine, a browser, and hosting for the apps the agent builds. win.sh watches the accounts you connect around the clock, proposes the next move, and acts inside an authority matrix you set.
How do Duet and win.sh compare?
| Dimension | Duet | win.sh |
|---|---|---|
| Core abstraction | An agent you teach once and share with each client | A 24/7 loop that monitors connected business accounts and acts |
| Human role | Teach the agent your method, share it, and keep improving it | Set rules per type of work, then review decisions and receipts |
| Execution | A workspace per client with an always-on machine, browser, and app hosting | Cloud sandboxes connected to accounts the customer owns |
| Strongest fit | Consultants who want clients to use their method directly | Founders who want the business watched with little daily involvement |
| Recurring work | Clients talk to the shared agent whenever they need it | Runs continuously, with recurring workflows and a morning brief |
| Control | Your rules in the agent, role-based access, and audit logs | Approval gates, authority levels, a hard budget cap, and receipts |
| Main cost | Work happens in conversations with the agent, with less shared process state than a workflow | The loop acts before the brief, so much of the review happens afterwards |
| Commercial model | 7 days free, then $20, $50, or $200 a month with unlimited seats | Self-serve monthly budget from $50 to $10,000, no revenue share |
Prices and features were checked against each vendor's own pages on 6 October 2026.
Who does Duet fit?
Duet fits a consultant or agency that wants clients to use its method without booking time. You brief the agent the way you would brief a new hire, with your checklists, templates, and rules, then share it with each client. Their people sign in and talk to it, their data stays in their own workspace, and when you sharpen a skill once, every client has it the next time they ask.
A single shared agent carries less structure than a recurring process needs. A monthly client report handled through conversation keeps its history in messages instead of a process you can inspect, and a colleague cannot easily see which step went wrong or who approved what.
Who does win.sh fit?
win.sh fits founders who want work found as well as done. Instead of waiting to be asked, the loop watches connected accounts such as Stripe, Plausible, or Google Search Console, spots a change worth acting on, and proposes or takes the next step. Each morning a Telegram brief and the Decisions tab show what it did and what needs you.
You decide how much it may do alone for each type of work, and your approvals, edits, and rejections become rules over time. Spending stops at a hard monthly cap, receipts show the cost of each action, and win.sh takes no revenue share. The price of that initiative is that the loop often acts before you have looked.
Where does each product put human control?
Duet keeps control in what you teach the agent and in who can use it. Your rules shape every answer, role-based access controls who sees what, and audit logs record what happened, while the work itself happens in conversations with the agent.
win.sh moves control into standing policy. The authority matrix sets what each type of work may do alone, and approval gates hold spending, outreach, publishing, and sensitive changes. A useful test is a client complaint that arrives at midnight: ask what each product would do with it before you wake up.
Where does Task Machine fit?
Task Machine suits work that has outgrown a single shared agent. Several agents and people share tasks, and each recurring job runs as an explicit workflow with approval, question, and verifier steps, so a teammate can open any run and see where it stopped. Requests start in Chat, approvals wait in one Inbox, and Tasks show the state of every job.
A playbook installs the agents, workflows, and documents for one job, such as preparing a monthly client report, so you start from a working setup. If clients mainly need to ask your agent questions, a shared Duet agent is the simpler fit.
Which should you choose?
Choose Duet if you want to package your method as an agent and give each client their own workspace with it. Choose win.sh if you want your own business watched around the clock and prefer to review a morning brief. Choose Task Machine if recurring client work needs explicit steps, approvals, and a record you can inspect.