Why are the SERP examples for this keyword all category names instead of instructions?+
Because the market they serve is enterprise procurement, not solo practice. Every article in the first page of Google for 'example of business process automation' is written to convince a VP that a category (AP, onboarding, supply chain) can be automated at scale. The case studies are Unilever, Nike, DHL, BMW. There is no room in that format for a literal user-typed instruction, because the target reader is not the person who will type it. Clone inverts this. The examples on this page are written for the person doing the Monday work, so they are shown as the artifacts that person touches: the chat sentence, the markdown file, and the app screen.
What exactly sits under memory/rituals/ on my machine?+
Plain markdown files, one per ritual. The invoicing example lives at memory/rituals/invoicing.md. Kickoff at memory/rituals/kickoff.md. Follow-ups, status, and closeout each get their own file. Every file has the same informal shape: a schedule or trigger, an apps_this_month mapping per client, a rate or template section, and an actions block listing what Clone should do. There is no schema. If it reads like instructions to a human assistant, Clone can run it. Because it is markdown, you can git-diff it, grep it, or edit it on your phone.
Is the apps_this_month idea just a list, or does it really drive behavior per client?+
It drives behavior per client. Every ritual that touches a client asset looks up apps_this_month[client] and uses that entry to decide which window to open. In the invoicing example, acme resolves to quickbooks_desktop so Clone opens the Desktop window. Nexora resolves to google_sheets so Clone opens invoice-template.gsheet. Holloway resolves to clio so Clone opens the billing tab. The Computer Agent does not have a branch per tool. It opens the mapped window and reads what is on screen. Changing 'acme: quickbooks_desktop' to 'acme: freshbooks' redirects the next Monday run for that one client with no other edits.
How is Clone running these examples without API keys or OAuth?+
The Computer Agent layer in src/components/architecture.tsx lines 18-22 is described as 'Reads the screen, clicks, types, scrolls'. That is the layer that executes the examples on this page. It does not connect to a tool's API. It opens the tool's window the same way you do, reads the UI through accessibility APIs, moves the cursor to the button you would click, and types into the field you would fill. No developer portal, no OAuth scope, no admin approval. The only credentials in play are the ones already logged into the app on your Mac.
Where do the named clients like acme, nexora, holloway, stellar, delta, fenway come from?+
Those are the illustrative client keys from the invoicing ritual that ships as a worked example in the Clone source, visible on the invoicing-automation-software guide at cl0ne.ai/t/invoicing-automation-software. They are not real clients. They exist so the apps_this_month section has enough rows to demonstrate the mapping behavior (one client on QuickBooks Desktop, one on a Sheet, one on Clio, one on Wave, one on Airtable, one on FreshBooks). When you copy the ritual for your practice, you replace those keys with your own clients and the mapped tools with the windows you actually keep open.
Are these 5 examples the only rituals Clone supports, or just the starters?+
They are starters, picked to cover the consulting lifecycle in consulting-business-workflow.md (invoicing is Phase 6 ongoing operations, kickoff is Phase 3 step 4, follow-ups is Phase 2, status is Phase 4 weekly rhythm, closeout is Phase 5). Any instruction you would give a human assistant in plain English can become a ritual. A ritual is just a markdown file Clone reads before it runs. If you have a Tuesday 3pm content-repurposing sequence, a per-client quarterly business review, or a monthly 'chase unpaid invoices at Net+7, +14, +21' process, it is another file under memory/rituals/ with the same shape.
How does review and rollback work on these examples?+
Every ritual on this page ends by staging artifacts, not sending them. Drafts sit in Gmail > Drafts. Invoices sit as drafts in whichever billing tool. Calendly invites are queued. Drive folders are created but no links are shared. You review, you approve, and then the sends fire. If Clone typed the wrong rate into QuickBooks Desktop, you fix the draft in QuickBooks the same way you would fix any draft, then correct the rate in invoicing.md so next Monday does not repeat the mistake. Principle 4 in architecture.tsx lines 61-63 is explicit: 'Every action Clone takes is logged and reversible, preview drafts before they send, see every file it touched, roll back an entire morning of work with one click'.
How does this compare to Zapier, Make, or Workato examples for the same keyword?+
Zapier and Make examples of business process automation are trigger-plus-action recipes configured inside the vendor UI: 'when Gmail label X, create Trello card'. They require an API or pre-built connector on both ends. That excludes QuickBooks Desktop, most practice-management billing tools, and any custom Airtable or Sheet-based flow. Workato and Blue Prism examples are enterprise RPA workflows requiring a bot server, a license, and usually a consulting partner. Clone examples are a sentence, a markdown file, and whatever window you already have open. No connector, no bot server, no vendor UI.
How long does it take to run one of the 5 examples end-to-end?+
Ten minutes to the first real output, once you have the ritual file written. Breakdown: one minute to install Clone, two minutes to copy the ritual file from this page and edit 6 to 10 lines for your own clients, five minutes to run the sentence in chat and watch Clone drive the apps, one minute to approve the drafts, one minute to add 'schedule:' to the file so it fires automatically next time. After the first run, the ongoing cost is the run itself, which is 30 to 120 seconds per ritual per cycle.
What happens when an app UI changes, say QuickBooks moves a menu item?+
You edit the one line of the ritual that describes that click. The actions block is prose. If Clone was looking for 'Customers > Create Invoices' and QuickBooks renames it to 'Customers > New Invoice', the line in invoicing.md becomes 'click Customers > New Invoice'. Clone picks up the change on the next run. The fragility tradeoff is real compared to a locked API contract, but the upside is the surface works for every app, including the ones that never had an API to begin with. The invariant is: if you can describe the click to a human assistant, Clone can do the click.
Is there a canonical example file I can read end-to-end right now?+
Yes. The invoicing ritual is visible in the guide at cl0ne.ai/t/invoicing-automation-software, which shows the full memory/rituals/invoicing.md file with the apps_this_month mapping for 6 clients, the per-engagement rates, and the 7-step actions block. That page is paired with a Monday chat transcript (TerminalOutput on the page) showing the literal instruction and Clone's read-open-type-save-capture sequence. Reading it tells you everything you need to write your own ritual for invoicing, kickoff, status, or closeout. The same shape applies to all 5 examples on this page.
Does any client data leave my Mac when these examples run?+
No client data leaves your Mac. Principle 1 in architecture.tsx lines 46-50 is literal: 'Clone operates your desktop apps from your desktop, client files, emails, contracts, and transcripts never leave your computer, your engagements stay confidential by default'. When the invoicing example moves data from your Timely window to your QuickBooks Desktop window, the data does not hop through anyone else's cloud. The Planner layer may call a model to interpret your English instruction, but attached client data is not sent along with it. If you install Clone, pay for it, and then uninstall, the ritual files and the memory stay on your disk. There is no export because there was no cloud copy.