The vocabulary, defined.

Every term used across device farm and agency traffic operations, written plainly, with why each one matters when you are the person paying for it.

• Device farmA rack of physical phones driven by software rather than by hand, used to operate many social media accounts a
• Phone farmAnother name for a device farm. A set of physical handsets connected to a controlling computer and driven by a
• OFMOnlyFans management. An agency model where a company signs creators and runs the commercial side of their busi
• Traffic accountA social media account whose purpose is to drive viewers toward a creator's paid platform, rather than to buil
• Account warmingUsing a new account like a person would before asking anything of it. Browsing, watching and scrolling for sev
• Fan outTaking one piece of content or one instruction and distributing it across many accounts at once. One upload be
• Job queueThe list of work waiting to run across a fleet. Each job targets one account on one phone, and the queue handl
• Ban healingDetecting that an account has been banned and automatically creating, warming and scheduling a replacement on
• Device fingerprintThe set of signals a platform reads to identify what hardware an account is running on. Model, identifiers, se
• EmulatorSoftware that pretends to be a phone on a computer. Cheap to run and routinely detected, because the hardware
• Remote control toolSoftware that mirrors many phone screens onto one computer so a person can drive them. Useful, but a human sti
• Mobile proxyA proxy that routes traffic through carrier IP addresses, so accounts appear to be on ordinary mobile networks
• Control chipThe USB hardware that sits between the controlling computer and a phone, allowing software to tap, swipe and t
• Kernel serviceThe background service on the controlling machine that talks to the control chips and relays instructions to t
• AgentThe software on the controlling machine that connects to the control plane, receives jobs and reports results.
• Control planeThe server that holds accounts, schedules and job state, and hands work to agents. The part of a fleet that is
• Fleet healthA single score built from offline phones, device faults, banned accounts and failed jobs, used to spot problem
• Account healthA score for an individual account built from its status, verification, posting history and the phone it lives
• Warm up windowThe period between creating an account and its first promotional post. Usually five to seven days, during whic
• Link stickerAn Instagram story element that sends a viewer to an external page. The main mechanism traffic accounts use to
• HighlightA saved collection of Instagram stories pinned to a profile. Used by traffic accounts to keep a link reachable
• Creator rosterThe set of creators an agency represents. Fleet structure is usually built around it, since accounts, content
• Hosted fleetA fleet where the provider owns, racks and runs the hardware, and the agency operates it through a dashboard.
• In house fleetA fleet where the agency owns the phones and runs them on its own premises, under licence, with installation,
• ThroughputHow much the fleet produces in a period, normally posts per day. Calculated as phones multiplied by accounts p
• SlotOne account position on one phone. A fifty phone fleet at three accounts per phone has a hundred and fifty slo
• RetryAutomatic re-execution of a job that failed, usually because a phone locked up or an app hung. Removes the nee
• TemplateA saved multi-step routine, such as post a reel, then a story with a link sticker, then add it to a highlight.
• Batch operationSelecting many accounts or phones and acting on all of them at once, rather than repeating the same action per
• Audit logA record of who did what and when, including every time a stored credential was viewed. The control that makes
• Envelope encryptionStoring credentials encrypted under a key that is itself encrypted, so stored data is useless without the key
• ConcurrencyHow many phones are executing jobs at the same moment. The difference between a fleet that works at two phones
• Detection at scaleThe problem that behaviour which passes on a few accounts becomes a visible pattern across hundreds, because t
• RampGrowing a fleet in measured stages, ten phones then thirty then fifty, checking ban rates at each stage rather

Device farm

A rack of physical phones driven by software rather than by hand, used to operate many social media accounts at once. Each device carries its own hardware identity, so platforms see real phones.

The reason agencies end up here is arithmetic. A VA handles twenty to thirty accounts before quality drops, so three hundred accounts means ten people and a manager. A farm moves the constraint from people to phones, and phones do not need training, do not quit and do not sleep. The trade is that you take on hardware, heat, power and a network plan, which is why most agencies either license a built system or pay someone to host it.

Phone farm

Another name for a device farm. A set of physical handsets connected to a controlling computer and driven by automation software on a schedule.

Phone farm and device farm mean the same thing, and which one you hear depends on who you are talking to. Operators in OFM tend to say phone farm. Vendors tend to say device farm because it sounds less like a warehouse of handsets, which is precisely what it is. Neither term tells you whether the phones are automated or merely remote controlled, and that distinction matters far more than the name.

OFM

OnlyFans management. An agency model where a company signs creators and runs the commercial side of their business, usually traffic, chatting, pricing and retention, for a share of revenue.

Traffic is the constraint in almost every OFM agency, which is why the traffic team is usually the largest and the hardest to manage. Chatting converts what traffic delivers, and pricing and retention shape the lifetime value, but none of it starts without views. That is why device farms took hold in this niche before any other: the link between account volume and revenue is immediate and countable.

Traffic account

A social media account whose purpose is to drive viewers toward a creator's paid platform, rather than to build a brand of its own. Agencies run many of them at once.

Traffic accounts are disposable by design, which changes how you treat them. You are not building a brand, so follower quality matters less than reach, and a ban is an operating cost rather than a disaster. The number that decides your economics is not how long an account survives but how quickly a dead slot is replaced and producing again.

Account warming

Using a new account like a person would before asking anything of it. Browsing, watching and scrolling for several days with no posting and no links, so the account looks lived in rather than freshly made.

Warming is the cheapest intervention available and the one most often skipped, because it costs a week of output per account and nobody wants to pay a person to scroll. That is exactly why it gets automated first. The failure mode is warming that is technically present but identical across every account, which creates a pattern rather than removing one.

Fan out

Taking one piece of content or one instruction and distributing it across many accounts at once. One upload becomes one job per phone.

Fan out is the difference between an operator and a fleet. Posting one reel to fifty accounts by hand is fifty actions with fifty opportunities to get it wrong. Fanning it out is one action that becomes fifty jobs, each retried independently if a phone hangs. It is also where scheduling variation has to be introduced, since fifty identical posts at the same minute is a signal.

Job queue

The list of work waiting to run across a fleet. Each job targets one account on one phone, and the queue handles ordering, retries and failures without an operator.

The queue is the part of a fleet that separates a demo from a system. At ten phones you can get away without one. At two hundred, one stuck device will stall everything behind it unless the queue leases work, times out, retries and trips a circuit breaker on a phone that keeps failing. Most in house builds stall here rather than at driving a phone.

Ban healing

Detecting that an account has been banned and automatically creating, warming and scheduling a replacement on the same phone, so the slot keeps producing.

This is the single feature that most changes agency economics, because it attacks downtime rather than ban rate. Manually, a banned slot stays dead three to seven days while somebody notices, rebuilds and rewarms. Automatically, it is hours. Across a hundred accounts with a steady ban rate, that difference is a large share of your total output.

Device fingerprint

The set of signals a platform reads to identify what hardware an account is running on. Model, identifiers, sensors, graphics and behaviour. A real phone produces a coherent one. An emulator has to fabricate it.

Understanding fingerprinting is what pushes operators from emulators to hardware. A platform does not need to catch you doing something wrong if it can simply tell that you are not running on a phone. Real devices pass this check for free, which is the entire argument for physical fleets and the reason they cost more.

Emulator

Software that pretends to be a phone on a computer. Cheap to run and routinely detected, because the hardware signals a real device produces are difficult to fake convincingly.

Emulators look unbeatable on cost per instance and lose badly on cost per surviving account. An instance is nearly free and an account that dies before it has built any reach is worth nothing. Agencies usually discover this after burning several hundred accounts, which is an expensive way to learn a fingerprinting lesson.

Remote control tool

Software that mirrors many phone screens onto one computer so a person can drive them. Useful, but a human still performs every action, so headcount does not fall.

The marketing in this category deliberately blurs remote control and automation, so test any tool with three questions. Can it create an account with nobody watching? Can it warm one? Can it replace a banned one? If the answer is no, you are buying a faster operator, not fewer operators, and your headcount will not change.

Mobile proxy

A proxy that routes traffic through carrier IP addresses, so accounts appear to be on ordinary mobile networks. Typically one proxy covers a group of phones.

Network design is usually decided too late. Twenty accounts behind one office connection are visibly one operation, and retrofitting proxies onto a built rack is more expensive and more annoying than planning them. Proxies sit between shared Wi-Fi and a SIM per phone on both cost and durability, and one proxy typically covers a group of devices.

Control chip

The USB hardware that sits between the controlling computer and a phone, allowing software to tap, swipe and type on the handset.

Chips are sold openly and are not proprietary to any platform, which is worth knowing because some vendors imply otherwise. They also fail, so spares are not optional. The chip is the cheap part of the stack and the part most likely to need physical attention, which is one practical argument for letting someone else host the rack.

Kernel service

The background service on the controlling machine that talks to the control chips and relays instructions to the phones. In common builds it runs on Windows only.

The Windows requirement catches people out constantly, usually after they have already bought a Mac to drive the fleet. The kernel service that talks to the chips does not run on macOS. Any hardware plan that starts with Apple hardware on the controlling side has to be redone, and it is better to find that out before purchase than after.

Agent

The software on the controlling machine that connects to the control plane, receives jobs and reports results. The link between the cloud and the phones.

The agent is the piece that has to survive the real world: power cuts, reboots, apps updating overnight, a chip disconnecting at three in the morning. Running it in a Command Prompt window works in a demo and fails in production, because nobody is going to drive to the office to restart it. It needs to register as a service and come back on its own.

Control plane

The server that holds accounts, schedules and job state, and hands work to agents. The part of a fleet that is not in your office.

Splitting the control plane from the phones is what makes multiple sites and multiple customers possible, and it is also what lets an agency run a fleet it does not physically hold. The practical consequence for an agency owner is that your schedule, accounts and history do not live on a laptop in the corner of an office.

Fleet health

A single score built from offline phones, device faults, banned accounts and failed jobs, used to spot problems before they cost output.

A single score is useful precisely because nobody reads twelve dashboards. What makes it worth having is that it points at the cause rather than announcing a problem: which phones are offline, which have faults, which accounts went down, which jobs failed. Output leaks quietly, and a score is how you notice before the month is over.

Account health

A score for an individual account built from its status, verification, posting history and the phone it lives on, used to decide what to post and what to retire.

Scoring accounts individually is what lets you act in batches without being reckless. You can push more to healthy accounts, hold back new ones, and retire the ones that are clearly dying rather than feeding them content. Without a score, every account is treated the same and your best content lands on your weakest slots.

Warm up window

The period between creating an account and its first promotional post. Usually five to seven days, during which the account only consumes and lightly interacts.

Five to seven days is the figure that holds up, and the temptation is always to shorten it because the account is sitting idle. Shortening it is the most reliable way to raise your ban rate. On a fleet the cost of waiting is zero in labour terms, which removes the only real reason anyone ever rushed it.

Link sticker

An Instagram story element that sends a viewer to an external page. The main mechanism traffic accounts use to move viewers to a creator's paid platform.

The link sticker is the actual mechanism by which a traffic account produces revenue, which makes it the most sensitive action an account performs. Posting one too early, or on an account that has not warmed, is the clearest possible signal of intent. Most sensible schedules hold links back until well after the first ordinary posts.

Highlight

A saved collection of Instagram stories pinned to a profile. Used by traffic accounts to keep a link reachable after the story expires.

Highlights matter because stories expire and traffic does not arrive on a schedule. Pinning a link-bearing story to the profile keeps the route open for anyone who finds the account days later, which is a meaningful share of total conversions on accounts that catch any reach at all.

Creator roster

The set of creators an agency represents. Fleet structure is usually built around it, since accounts, content and manager access are all organised per creator.

The roster is usually the right organising principle for a fleet, because it maps onto how an agency already works. Phones group by creator, managers are scoped to the creators they handle, and content is organised the same way. Structuring a fleet by anything else tends to create a translation layer that somebody has to maintain by hand.

Hosted fleet

A fleet where the provider owns, racks and runs the hardware, and the agency operates it through a dashboard. No hardware in the agency's office.

Hosted is the faster route and the one with no capital outlay, since the hardware already exists and is already racked. What you give up is physical control of the devices. For most agencies that is a straightforward trade, because the hardware was never the thing they wanted to own.

In house fleet

A fleet where the agency owns the phones and runs them on its own premises, under licence, with installation, training and support from the provider.

In house suits agencies that want the capability as an asset rather than a subscription, and that have somewhere to put a rack. It takes longer to start because hardware has to be specified, sourced and installed, and it requires at least one person on your side who is comfortable owning the operational side.

Throughput

How much the fleet produces in a period, normally posts per day. Calculated as phones multiplied by accounts per phone multiplied by posts per account.

Throughput is the number to design backwards from. Fifty phones at three accounts at three posts is four hundred and fifty posts a day, and that is four hundred and fifty pieces of content you have to supply. Content supply, not phone count, is the constraint most agencies hit first and plan for last.

Slot

One account position on one phone. A fifty phone fleet at three accounts per phone has a hundred and fifty slots. Bans take slots out of production temporarily.

Thinking in slots rather than accounts clarifies the economics. A slot is a unit of production capacity that is sometimes occupied by a healthy account and sometimes empty because of a ban. Your real output is slots multiplied by the share of time they are occupied, which is why replacement speed matters as much as ban rate.

Retry

Automatic re-execution of a job that failed, usually because a phone locked up or an app hung. Removes the need for someone to notice and act.

Retries exist because phones fail in boring ways: an app hangs, a dialog appears, the screen does not load. Without automatic retry, every one of those becomes a human noticing something and redoing it. With retry, the vast majority never reach a person at all, and the ones that do are the genuinely broken cases worth looking at.

Template

A saved multi-step routine, such as post a reel, then a story with a link sticker, then add it to a highlight. Run across any set of accounts in one action.

Templates are what turn a fleet from a posting tool into an operation. A real routine is rarely one action: post the reel, then the story with the link sticker, then add it to the highlight. Saving that as one named thing means it runs identically every time and a new team member cannot do it in the wrong order.

Batch operation

Selecting many accounts or phones and acting on all of them at once, rather than repeating the same action per item.

Batch actions are the difference between managing fifty accounts and managing a fleet. The important design detail is that batching must respect account health, so that a single bulk action does not push identical content to every account at the same moment, which is the fastest way to turn a convenience feature into a detection problem.

Audit log

A record of who did what and when, including every time a stored credential was viewed. The control that makes shared access safe.

Audit logs matter most in businesses with turnover, and OFM has plenty. When a manager leaves, the question is not only what they can still access but what they looked at while they were there. A log of credential views turns an uncomfortable unknown into a record you can read.

Envelope encryption

Storing credentials encrypted under a key that is itself encrypted, so stored data is useless without the key belonging to the account holder.

The practical effect is that a copy of the database is not a copy of your logins. For an agency this is the difference between a technical incident and a disaster involving every creator you represent. It is also what makes it reasonable to give managers operational access without giving them the credentials themselves.

Concurrency

How many phones are executing jobs at the same moment. The difference between a fleet that works at two phones and one that works at two hundred.

Concurrency is where most self built fleets break. Driving one phone is a week of work. Driving two hundred at once introduces contention for USB bandwidth, for the controlling machine, for network, and for the queue itself. Two phones working proves the automation logic and proves nothing about two hundred.

Detection at scale

The problem that behaviour which passes on a few accounts becomes a visible pattern across hundreds, because the similarity itself is the signal.

This is the counterintuitive part of running volume. Behaviour that is perfectly safe on five accounts becomes a signature on five hundred, because the similarity between them is itself the evidence. It is why randomisation has to be genuine rather than a fixed offset, and why fleets are ramped in stages rather than deployed at full size.

Ramp

Growing a fleet in measured stages, ten phones then thirty then fifty, checking ban rates at each stage rather than deploying everything at once.

Ramping exists because ban rate is the only number that tells you whether your setup actually works, and you can only learn it by running. Ten phones, then thirty, then fifty, measuring at each step, costs you a few weeks. Deploying two hundred at once and getting it wrong costs you two hundred accounts and the time to rebuild them.