Business process automation platforms
Ranked by a number that is also their ceiling.
Every page-one guide ranks BPA platforms by connector count. Zapier 8,000+. Make 2,000+. Power Automate's certified catalog. n8n's node library. The count is not the feature. It is the cap of what any API-based platform can ever automate. Clone ships with zero connectors because it drives your screen, not your APIs. This page makes the claim checkable.
The platforms the SERP is ranking
Eighteen names, one shared ranking axis.
Here are the vendors that show up on every "best business process automation platforms" listicle, roughly in the order the articles present them. They differ in canvas shape, pricing model, hosting, and target buyer. They do not differ in the shape of their ranking axis. Each one markets a connector count, and each one has a ceiling at that count.
The claim that makes this page uncopyable
Clone ships zero connectors. You can verify it in your terminal.
The Clone website and product live in one codebase. Clone it, open the src folder, and run the commands below. There is no connectors directory, no integrations directory, no api-registry directory, and no file whose name contains either word. The single string connector does not appear in any integration file because there is no integration file. The execution layer is the Clone Computer Agent, and it is six words in architecture.tsx.
The file that replaces the connector catalog
architecture.tsx, lines 14-42, verbatim.
What a BPA platform stores as a catalog of connector modules, Clone collapses into one line in one file: the Clone Computer Agent reads the screen, clicks, types, scrolls. Here is the literal excerpt from the repo. There is nothing hidden under it. No secret connector map, no shimmed SDK layer, no service bindings file. The agent is the catalog.
The numbers
Zero, eight thousand, forty-nine, twenty-one.
The four numbers that matter on this page, and one SERP number sitting in the middle of them for contrast.
Why the SERP ranks on the wrong axis
Connector count is convenient for the writer, not useful for the buyer.
The ranking axis every SERP article uses is the one that makes the article easiest to write. It is a single integer per vendor. It ranks cleanly. It is easy to cite. It has none of the properties that would make it useful for a small firm choosing a platform.
Connector count compares neatly
A single big number per vendor turns a sprawling product into a league table. It is editorially convenient, which is why Gartner Peer Insights, Fortune Business Insights, Cflow, and Moxo all lead with it. Convenience for the writer is not the same as usefulness to the buyer.
The number hides the depth
Saying a platform covers 8,000 apps does not tell you which actions are exposed, which are premium, which are rate-limited, and which are read-only. The number compares totals, but buyers will hit their problem on the depth axis, not the breadth axis.
The number omits the long tail
Every small consulting practice has at least one app that is not on any catalog. A portal an insurer set up in 2011. A white-labeled HubSpot variant. A Google Sheet that is really the CRM. The top 100 apps on every BPA vendor's catalog are shared. The long tail is not, and the long tail is where real work lives.
The number omits screen-driven operators
There is a whole category that does not live inside a catalog: screen-driven operators, instructed in English, that work with any app a human can see. Clone is one. Adept-style agents are another. Because this category does not compete on catalog count, BPA listicles do not know which column to put it in, so they leave it out.
The catalog is the ceiling, four ways
Every ceiling above the Clone Computer Agent.
Being an API-based platform is not one constraint, it is several. Each of the cards below is a separate ceiling the buyer hits, and each of them is invisible on a listicle that only compares total connector counts.
Your app is not in the catalog
Custom internal tools, an older QuickBooks Desktop install, a niche vertical CRM a client insists on, a newly launched SaaS still in beta. None of these ship with a Zapier app or a Power Automate connector on day one. On API-based platforms, this puts the work on the wrong side of the ceiling: you wait for a connector, or you build one yourself. Clone operates all of them identically because it operates the screen.
The catalog does not cover every action
A platform that has your app may only expose 6 of its 40 actions through the connector. Zapier supports triggering on a QuickBooks Online invoice payment, but not editing a class assignment. Power Automate supports reading Dynamics records, but not every custom entity. Connector breadth and connector depth are different ceilings, and both apply.
APIs deprecate. Screens do not
When a vendor retires a v2 endpoint in favor of v3, every API-based platform has to rewrite its connector. Your workflow breaks silently until the connector catches up. A visual workflow you built in March 2024 stops running in August when Twitter v1.1 goes away. The screen a human uses to do the same thing, however, is still the same screen.
The catalog also charges extra
Premium connectors, tier-gated operations, per-task metering tied to which app is in the chain. Power Automate sells its Premium Connectors in a separate SKU. Workato charges per recipe run plus per connector. The catalog is also a pricing surface, not only a capability surface.
Clone runs under the API layer
The Clone Computer Agent is beneath the network. It interacts with the pixels a human sees. Every ceiling above (catalog coverage, connector depth, API version churn, premium SKU gating) is irrelevant to it, because nothing the agent does requires an integration contract between Clone and the target vendor.
Long-tail apps in, finished work out
Apps that no BPA platform has a connector for, all feeding the same hub.
On the left: the awkward apps a solo practice really uses. On the right: the work a solo practice really needs done. In the middle: the Clone Computer Agent, which bridges them without a single connector file existing.
long-tail apps → screen-driven hub → finished work
How a run actually goes
From an English sentence to drafted invoices, without an API in the middle.
Here is what one request looks like, layer by layer. The architecture file is the skeleton; these are the five beats a real run lands on.
You type a sentence
The user-visible input is a chat window. You say what you want the way you would ask a junior employee, for example Issue all my client invoices for last week's hours. No trigger dropdown. No branch modal. No OAuth screen to pre-wire.
Clone Planner chooses apps
The Planner layer, defined in architecture.tsx line 13-17, reads the intent and figures out which of the apps already open on your Mac belong in the job. QuickBooks Desktop, QuickBooks Online, FreshBooks, Xero, a custom invoicing webapp at an internal URL. They are all eligible because none of them need to be pre-registered.
Clone Computer Agent drives the screen
This is the layer that does the work. It looks at the pixels on your display, reads text, locates buttons, types into fields, scrolls lists, clicks Save. It operates the app the way you do. No API call is made to anything it does not strictly need, and the agent never depends on an app being in a connector catalog.
Clone Memory captures the pattern
What you asked, which apps got touched, which templates were used, and every step the Computer Agent took are written to ~/.clone/memory/. Next time you ask for the same thing, the steps are recalled. The memory is markdown you can grep and diff in git.
Your business updates, in its real tools
Invoices live in the invoicing tool you already paid for. Emails are in Gmail. Contact updates are in HubSpot. Nothing was shuttled through an intermediate orchestration cloud, and nothing got copied into a vendor database. Your stack is unchanged, but the work got done.
The one-sentence rule of thumb
If a BPA platform can only automate apps in its catalog, it cannot automate yours past its ceiling.
Pull up your last month of client work. Make a list of every application you touched. Any of them not on the major BPA platforms' catalogs is work the platforms cannot cover. Clone's answer to that list is the same regardless of what is on it: the Computer Agent opens the app, the agent uses the app, the agent closes the app.
That is the whole difference. Not a pricing cut, not a faster canvas, not a better AI copilot inside the same canvas. A category below the API, instructed in English.
“The connector catalog is the cap of what any API-based BPA platform will ever automate. Clone has no catalog because it does not call APIs to do the work.”
Clone, architecture.tsx
Row-by-row, how a screen-driven operator differs from a catalog-based platform
Eight axes that are not connector count
The row every SERP listicle ranks on (total connectors) is missing, on purpose. These are the axes where the buying decision actually lives.
| Feature | Catalog-based BPA platforms | Clone |
|---|---|---|
| Catalog size | Fixed: 500 to 8,000 apps, per vendor | Unbounded: any macOS app a human can see |
| Works with apps not in the catalog | No, unless the buyer writes a custom connector against an exposed API | Yes. The Computer Agent operates the UI, not the API |
| Breaks when an API deprecates | Yes. A v2 to v3 migration is a maintenance event | No. The same buttons, fields, and pages keep working |
| Covers premium or gated endpoints | Only via premium connector SKUs, often with per-operation metering | Not applicable. The agent uses whatever the user's screen shows |
| How the buyer authors work | Visual canvas or node graph. Trigger, action, branch, filter modals | Chat window and a folder of markdown at ~/.clone/memory/ |
| How the work is debugged | Replay the run in the canvas, inspect step logs inside the vendor UI | Open the markdown ritual file and the screen log in any editor |
| What happens when a client changes tools | Rebuild the workflow against the new connector, re-auth, remap every field | Edit a line in the ritual file. The agent uses the new app's screen |
| Where the workflow graph lives | In the vendor's cloud. Portable only through a vendor-specific export | In ~/.clone/memory/ on your Mac. Grep-able, diff-able in git |
Four principles, not a feature list
Plain English in, apps touched out. No vendor allowlist. Portable files. Runs on your Mac.
These are not marketing principles. They are the constraints the architecture.tsx file and the ritual file format enforce. A BPA platform could promise any of them in a blog post; none of the catalog-based ones can structurally deliver all four.
Plain English in, apps touched out
The input is a sentence you could say to a new hire. The output is work completed inside the apps you already use. The layer that bridges the two is the Clone Computer Agent, and it does not call APIs to do it.
No vendor allowlist for the target apps
Because the agent operates at the OS level, there is no list of supported apps to maintain. This is the architectural reason the repo does not contain a connectors directory: there is nothing to list.
Portable, readable workflow definitions
What API-based platforms store as a proprietary JSON graph, Clone stores as a markdown file in ~/.clone/memory/. You can version it in git, grep it with your shell, and port it to any other macOS by copying the folder.
Your data does not leave your Mac
API-based orchestration platforms proxy every payload through their cloud. Clone's agent runs locally. Client files, emails, and transcripts never pass through a vendor's data plane, which is the reason the architecture.tsx principles section leads with Runs on your machine.
When Clone is the right choice
Four shapes of practice Clone is built for.
Your stack has one or more custom or legacy apps
If any part of your process runs in software that does not ship a connector on the major BPA platforms, Clone is the natural fit. Because the agent uses the same screen you do, the custom part is not special.
You do not have a full-time automation admin
Canvas-based platforms assume a dedicated owner who knows the DSL, the connector auth quirks, and the error modes. Solo consultants and 2-5 person firms rarely have that. Clone's surface area is a chat window, so the owner is whoever already does the work.
Your clients change tools more than once a year
If the invoicing tool, CRM, or client portal you operate changes across clients, rewiring a canvas every time is expensive. A ritual file is one sentence to edit.
You need to keep data on your machine
Regulated work (legal, healthcare, finance) often rules out sending documents and transcripts through a third-party orchestration cloud. Clone's Computer Agent keeps everything on the Mac.
When a platform is the right choice
Four shapes of work where Zapier, Power Automate, UiPath, or Appian beat Clone.
Honest pages tell you when to buy the other thing. Clone is not the right answer for every shape of automation work. These are the four shapes where it is not.
You orchestrate many SaaS APIs at high volume
If the work is tens of thousands of API calls per day across a defined list of cloud apps, a platform built on those APIs is usually cheaper and more auditable. Zapier, Workato, and Tray.io are strong here.
You live inside Microsoft 365
Teams, SharePoint, Dynamics, and the rest of Microsoft's estate are Power Automate's home turf. Connectors are first-class, and most enterprises already have licensing.
You need a BPMN model an auditor can read
Heavily regulated workflows benefit from a formal BPMN diagram that a compliance team can review and sign off. Appian, Camunda, and Bizagi are the deep end of that need. Clone does not produce a BPMN artifact; if your auditor wants one, buy one of those.
You employ a full-time automation team
If the automation admin is a real role (RPA developer, Zap builder, workflow architect), the canvas investment pays back. Clone is built for the case where that role does not exist.
“I had a client demand we automate their internal claims portal. It was a Rails app they built in 2015 with no public API, no Zapier app, and definitely no Power Automate connector. Our ops team had just signed a new Workato contract and they said they could build a custom connector, eight weeks, five-figure budget. I installed Clone that afternoon, described the weekly claims pull in plain English, and the ritual was running the next morning. The thing that sold me was that Clone did not care the app was custom. It just used the app.”
The thing the SERP is not telling you
The best business process automation platform for your practice is, structurally, the one that does not need your apps to be in a catalog.
The SERP lists 10-20 platforms every year. Ninety percent of them are catalog-based, with the rest split between canvas RPA and enterprise BPM. None of the listicles surface the category Clone lives in, because it does not have a number the listicle can rank it on.
Your job as a buyer is to notice that the ranking axis is chosen, not given.
Book a 30-minute call
Bring the three BPA platforms you are evaluating.
We run through each one on the axes that actually matter to a small practice: long-tail app coverage, behavior when your client changes tools, what breaks when an API deprecates, and whether the workflow definition is portable. If Clone wins, we show you the ritual file. If a platform wins, we point you at it.
Book a 30-minute callYour current BPA platform has a ceiling. We find yours in 20 minutes.
Twenty minutes together. Tell us which BPA platform you are on; we name the number that is its ceiling and show where Clone picks up.
Business process automation platforms, the axis-is-wrong edition
Why do business process automation platforms all get ranked by connector count on page one of Google?
Connector count is editorially tidy. It collapses a sprawling product into one number per vendor, which turns a buyer's guide into a league table. It also happens to be the easiest field to pull from each vendor's marketing site. None of this makes it the right axis to buy on. For most solo consultants and small firms, depth per app matters more than breadth across apps, and whether the specific apps they already use are in the catalog matters more than the total count.
Is Clone a business process automation platform in the Zapier, Make, Power Automate sense?
No, not in the sense the listicles use the word. Clone does not expose a workflow canvas, it does not maintain a connector catalog, and it does not charge per API operation. Its execution layer is the Clone Computer Agent, which operates macOS apps at the UI level. It belongs to a different category, screen-driven operators instructed in plain English, which the listicles mostly ignore because it does not rank on their chosen axis.
What happens if my invoicing tool or CRM does not have a connector anywhere?
That is the case Clone is built for. The agent uses the screen a human would use. If you can open the tool, log in, and click the right buttons yourself, Clone can too. Custom internal tools, older desktop apps like QuickBooks Desktop, and niche vertical SaaS without public APIs are all inside Clone's scope because nothing in its architecture depends on a connector.
How can I verify the no-connector claim in the Clone repo myself?
Clone's website is an open codebase. Clone the website repo, then run find src -type d -name 'connectors' -o -name 'integrations' -o -name 'api-registry'. The command returns nothing. Run grep -rn connector src and you will see that the word connector shows up only in marketing prose comparing Clone against catalog-based platforms, never as an actual integration file.
Does screen-driven automation mean Clone is slower than an API-based platform?
For high-volume structured flows across cloud APIs, yes, a native API platform will be faster per operation. For the mixed kind of work a consulting practice runs (invoicing 6 retainer clients each Monday, drafting status emails, updating a CRM, filing a report), Clone is faster end-to-end because there is no setup phase. The first ritual is running within 10 minutes, whereas a canvas-based platform takes days to wire the equivalent coverage, if the connectors exist at all.
When does a platform like Power Automate or UiPath make more sense than Clone?
When you have a dedicated automation team, when the workflows are almost entirely Microsoft-centric or enterprise SaaS with stable APIs, when you need a BPMN artifact an auditor can sign, or when the volume is high enough that per-operation metering is cheaper than a flat monthly fee. Clone is built for the case where none of those hold: a solo or small practice, a mixed stack with at least one awkward app, and a preference for plain English over a canvas.
Where does Clone fit alongside the enterprise RPA category (UiPath, Automation Anywhere, Blue Prism)?
Enterprise RPA is the closest cousin: they also drive UIs rather than APIs. The difference is the authoring surface. RPA platforms still require a workflow designer, a recording session, a published package, and usually a license server. Clone replaces all of that with a chat window and a folder of markdown rituals. It is, roughly, the shape RPA could take if you removed the canvas and the enterprise deployment layer.
Where is the file you keep referring to, and can I read it?
The file is src/components/architecture.tsx in the cl0ne.ai website repo. Lines 14 through 42 define the six layers the product is composed of, including the Clone Computer Agent layer that says Reads the screen, clicks, types, scrolls. Lines 44 through 65 define the four founding principles. The whole file is 159 lines. That file is the reason this page is not a list of 20 vendors with stars next to them.
Each one picks a different uncopyable property of Clone and defends it from a file in the repo.
Adjacent pages on the same thesis
A Business Process Automation Tool With No Graph Editor. On Purpose.
Zero workflow canvases in the product. The user-visible input is a chat window on top of a folder of markdown. The structural argument for why the canvas is the problem, not the feature.
Advantages of Business Process Automation, Grounded in a File You Can Open
Every SERP article recites 8-10 advantages without naming a mechanism. Clone's architecture.tsx names 3 layers and 4 principles. Each advantage maps to a layer plus a principle, at a specific line number.
Business Process Automation Service For Solo Consultants
What changes when the service is something that operates your screen rather than configures a workflow for you. A different engagement model, a different unit of delivery.
Zero connectors. Every Mac app. $49/mo, 21-day free trial.
Book a call