A Claude iPhone farm lead generation setup runs an AI agent on a group of iPhones so the phones read Instagram, LinkedIn, Twitter and Reddit themselves, then send outreach from the same devices. Kevin built one to replace paid scraping and messaging tools, and the walkthrough is on his channel.
That is the whole idea in one line: the model does the reading and the writing, the phone does the clicking, and you stop paying four separate SaaS bills to move data between them. The video description lists five jobs the farm takes over: running Claude Haiku 5.5 on the phones, pulling leads from Instagram, LinkedIn, Twitter and Reddit, sending LinkedIn connection requests every day, running one iMessage campaign per client on its own phone, and scaling that until the recurring software cost drops.
This post is about Kevin’s own open-source project and the iPhone farm video. It is his build, not a vendor product tour, and the source is here: the iPhone farm walkthrough. Watch the rest of the builds on the KevBuildsApps YouTube channel.
You (one prompt, one goal)
|
v
+---------------------+
| Agent loop |
| plan, read, act |
+----------+----------+
|
+---------+---------+---------+
| | | |
v v v v
Phone 1 Phone 2 Phone 3 Phone N
Insta LinkedIn Twitter Reddit
scrape invites leads threads
| | | |
+---------+---------+---------+
|
v
Leads sheet + outreach sent
(from the phones, not an API bill)
What Is a Claude iPhone Farm and How Does It Work?
A Claude iPhone farm is a set of iPhones that each run an AI agent under a frontier model, doing lead research and outreach inside the normal apps instead of through a paid scraper or a per seat outreach tool.
The loop is simple. You give the agent a target, say “find 40 e-commerce founders on Instagram who posted about subscriptions in the last month.” The agent reads the feed on the phone, saves the profiles that match, drafts a first message, and sends it. The phone is logged into Instagram the way a person would be, because it is a person’s account, on a device.
Three things make this work in 2026 that did not work a couple of years ago:
- Small frontier models are good enough. Claude Haiku 5.5 is in the current Claude lineup, and the video’s framing is that a fast model handles classification, matching and message drafting for pennies.
- Phones are the cheapest always-on computer you already own. A used iPhone costs less than a year of most lead scraping seats, and it comes with the apps.
- The agent can be driven by Claude Code. You write the loop once, and each phone runs its slice of it. Kevin shows the same pattern in building a social media agent with Claude Code.
What it is not: a magic way to avoid rate limits forever or a promise that no account ever gets flagged. It is a cost decision. You replace a metered scraping API with hardware you control and a model you pay per token.
How Claude Haiku 5.5 on iPhones Replaces Expensive Lead Gen Software
The expensive part of lead gen software is rarely the scraping. It is the seats, the credits, the enrichment add-ons and the outreach module sold separately.
Put the model on the phone and the pricing model changes. You pay for tokens and for the phone. Here is the shift, framed as what the farm does, not as a claim about any specific vendor:
| Job | Typical paid tool shape | What the iPhone farm does |
|---|---|---|
| Lead discovery | Monthly seat plus credit packs | Phone reads the platform in the app, agent filters |
| Enrichment | Per contact, per email | Agent reads the profile page and writes fields to a sheet |
| Outreach | Separate outreach product | Agent taps through the connection request on the phone |
| Messaging | Per message or per campaign | One phone per client campaign, no send fees |
A worked example of the prompt Kevin uses the pattern for:
You are a lead researcher. Open Instagram on this phone.
Search the hashtag #dtcbrand. For the 30 most recent posts,
open each profile. Keep founders whose bio mentions
"shop" or "store" and who posted in the last 14 days.
Write name, handle, follower count and the post link to
leads.csv. Do not message anyone yet.That job costs a few cents of model tokens and a few minutes of phone time. The same research in a seat-based scraper usually costs a monthly fee whether you use it or not. If you want the account-management side of this handled by Creator OS instead of raw phones, the agents docs cover how an MCP-capable agent runs the accounts.
iPhone Device Farm Setup: Hardware, Software, and Costs
A working farm needs four things: phones, a way to run scripts on them, a model endpoint, and somewhere to put the leads.
Kevin’s video walks through the setup he actually uses. The parts that matter most:
- Phones. Refurbished iPhones are fine. Each phone gets its own Apple ID and its own app logins. One phone per account per platform keeps things clean.
- Control layer. Something has to press the buttons. That is the piece the video spends the most time on, because it is where reliability lives.
- Model. Claude Haiku 5.5 for the high volume work (classify, match, draft), a bigger model when a task needs judgment. You can point the agent at a logged-in Claude Code session or an API key, the same way the open source iPhone farm writeup describes.
- Storage. A Google Sheet or a CSV per campaign. Nothing fancy. The agent appends, you read.
Budget roughly: a used iPhone, a power strip and a shelf, and pennies per hundred model calls. Compare that to a stack of four subscriptions at typical lead gen pricing and the monthly line usually drops at small scale.
Two setup rules that save you a week:
- Charge and network all phones from one hub so a reboot does not strand a device.
- Name every phone by the account it holds (“li-kevin”), not by number. You will forget which is which.
Replacing the Scraper: Reading Instagram, LinkedIn, Twitter, and Reddit for Pennies
The farm replaces a lead scraping tool with the phone reading the same four networks itself: Instagram, LinkedIn, Twitter and Reddit.
The pattern per network is the same shape with different buttons:
| Network | What the agent reads | What it saves |
|---|---|---|
| Hashtag feeds, profiles, recent posts | Handle, bio, follower count, post link | |
| Search results, profile pages | Name, title, company, profile URL | |
| Keyword search, replies to a post | Handle, bio, the tweet that matched | |
| Subreddit feeds, comment threads | Username, subreddit, the comment text |
Reddit is the one most people skip and the one that pays. If you would rather not run it from a phone, Creator OS has a Reddit group of endpoints (Reddit API docs) plus research tools like reddit_list_subreddits, reddit_search and reddit_subreddit_rules in the MCP server, so an agent can find threads and post into them without scraping. Same for the other networks: the hosted MCP server exposes about 160 tools covering posts, analytics, comments and DMs.
The economics are the point. A scraped lead costs a few cents of tokens plus the phone’s share of the hardware. That is the “pennies” in the video title.
Dropping the Outreach Tool: Automated LinkedIn Outreach from Your Phone
The farm replaces a separate outreach product with connection requests sent from the phone, every day, on a schedule the agent owns.
The loop is deliberately boring, which is why it works:
For each row in leads.csv where status is empty:
Open LinkedIn. Search the person by name and company.
If the profile matches, send a connection request with
the note from notes.txt. Set status to "requested".
Stop after 20 requests today. Log every action.Twenty a day is a pace a person could keep, and the agent keeps it every day without you opening the app. The agent also reads the inbox on the same phone and replies to anyone who accepts, which is the half most outreach tools leave to you.
If LinkedIn is one channel of a wider operation, Creator OS posts to LinkedIn through the same API as everything else, with first comment, organization pages and audience targeting on page posts. The post options docs list the per-platform settings. Use the phone farm for the outreach that has no API, use the API for the publishing that does, and skip the per seat pricing on both.
iMessage Marketing on Each iPhone: Running Campaigns for E-commerce Clients
The third replacement is iMessage marketing. Each e-commerce client campaign gets its own iPhone, so every client’s messages go out from that client’s own number.
Why one phone per client matters:
- Conversations stay separate. A customer replying to your skincare client never lands in your supplement client’s thread.
- A problem on one phone does not touch the others.
- You can report per client without filtering a merged inbox.
A campaign on that phone looks like a list and a template:
Campaign: restock_alert (client: skincare)
List: customers.csv, opted_in = true
Message: "Hi {first}, the serum is back in stock
today. Want me to hold one for you?"
Send between 10am and 4pm local. Stop at 60 per hour.
Reply to anyone who answers within 5 minutes.Only message people who opted in. That is not a legal footnote, it is the reason the channel keeps working: an iMessage thread is a personal channel, and the agent should treat it like one. If a client wants WhatsApp instead, Creator OS handles WhatsApp Business messaging with approved templates outside the 24 hour window, and Meta bills template deliveries to the client’s own WhatsApp Business account, so you are not eating the send cost.
Scaling Your iPhone Farm Without Losing the Thread
The savings come from replacing a stack of paid tools with hardware and tokens. Scale the farm in small steps and the savings compound instead of breaking.
How to scale without losing the thread:
- Add phones one at a time. One new phone, one new account, run for a week. Two at once and you will not know which broke.
- Keep one leads file per campaign. When volume grows, split by network, not by phone.
- Watch token spend separately from hardware spend. Tokens scale with volume, phones scale in steps. Both are cheap, but you want to see which line is moving.
- Automate the reporting. A short daily summary from the agent (“42 leads, 19 requests sent, 6 replies, 1 booked call”) is the whole dashboard.
Then hand the publishing side to something built for it. Creator OS publishes, schedules, replies and reads analytics on 11 platforms plus Skool from the web app, the iOS app, a REST API, the CLI and a hosted MCP server. The same idea as the farm, minus the shelf of phones: an agent runs the accounts. If you already have leads from the farm, the content repurposing agent post shows how the same agent turns one clip into posts on every platform.
iPhone Farm GitHub and Open Source Resources (2026)
Kevin’s open-source work is on GitHub under kevinbadi, and the iPhone farm video is the place to see it applied. The three repositories worth knowing in 2026:
| Repo | What it is | Why it matters here |
|---|---|---|
| social-agents | Open-source agent harness for your socials on Creator OS | Interviews you about your brand, then posts, replies, runs automations and reports |
| creatoros-claude-plugin | Claude Code plugin for Creator OS | Adds the MCP server, campaign builder skills and a publish guard |
| open-edits | Claude Code skills that edit talking-head video | Turns the raw clip into the short you post after the leads land |
Social Agents is the closest thing to the farm’s brain in a repo. Kevin’s project starts with npm start creatoros social-agents, interviews you about your brand, then posts, replies to comments and DMs, runs automations and reports from a local dashboard. You bring your own model: a logged-in Claude Code session, an ANTHROPIC_API_KEY, or any Anthropic-compatible API. That is the same bring-your-own-model choice the phones make.
The plugin is the shortcut to the Creator OS side. Add the marketplace with /plugin marketplace add kevinbadi/creatoros-claude-plugin, install creatoros@creatoros, and configure it with your cos_live_ key. It adds a publish guard that asks Proceed or Cancel before anything is published, spent or deleted, which is the habit you want when an agent is touching live accounts.
For the iOS automation details, watch the full iPhone farm breakdown. For how Kevin builds these, the channel is youtube.com/@KevBuildsApps, and the launch video for Creator OS is here. The open source iPhone farm writeup on this blog covers the automation internals in more depth: open source iPhone farm iOS automation.
FAQ: Claude iPhone farm lead generation
What is an iPhone device farm and how does it work?
An iPhone device farm is a group of physical iPhones controlled by software so automation runs inside the real apps on each device. In Kevin’s build, each phone holds one account per platform and an AI agent reads feeds, collects leads and sends outreach on that phone. You run one agent loop, and each phone executes its share of the work.
Can I run Claude on an iPhone for lead generation?
Yes, in the sense the video uses it: the phone drives the app and a Claude model does the reading and writing for each step. Kevin’s setup keeps the model cheap by using a fast model like Claude Haiku 5.5 for the high volume classification and drafting, and the agent can be pointed at a logged-in Claude Code session or an API key. The phone is the hands, the model is the judgment.
How much does an iPhone farm cost to run?
The farm’s recurring cost is model tokens plus the electricity and shelf space for the phones, and the upfront cost is refurbished hardware. Check current pricing on any tool you are replacing against your own invoice before you commit. The phones cost less than most per seat scraping plans over a year.
Is there an iPhone farm GitHub repository with open source tools?
Kevin’s open-source repos are on GitHub under kevinbadi, and the closest to the farm’s brain is social-agents, the agent harness that runs your socials on Creator OS. The creatoros-claude-plugin repo adds the MCP server and campaign builder skills to Claude Code. Read the README of each before you install, because the install steps live there.
What about Claude skills and Claude reference for iOS automation?
The open-edits repo is where Kevin publishes Claude Code skills, in that case for editing talking-head video. For the iOS automation internals, the video walkthrough is the reference, along with our post on open source iPhone farm iOS automation. Creator OS itself ships skills through the CLI, and creatoros sync pulls the current set, including the Meta, Google, TikTok and LinkedIn campaign builders.
Get started
The farm handles the outreach. Creator OS handles the publishing, comments, DMs and analytics on the accounts you are growing, from a hosted MCP server, a CLI, a REST API and an iOS app.
Start at creatoros.ca/sign-up. The Creator Premium plan is $14.99/month or $59.99/year for one set of socials across 11 platforms plus Skool, and every plan includes API keys, the MCP server, the CLI and the agent skills. New subscribers start for $0 with a free trial and can cancel anytime.
Then watch the builds: the iPhone farm walkthrough and the Creator OS launch video on KevBuildsApps. The rest of the API and MCP details are at creatoros.ca/docs.
Get started for $0
Free trial for new subscribers. Cancel anytime.
