Reviewer: 6 years on phone farms Verdict score 6.4/10

I've Run Android Poker Bot Farms Since 2019.

If you searched "poker bot for android", you probably wanted an APK to sideload. An Android-based poker bot takes more than a sideloaded APK. I've been on a phone-farm floor at 2am replacing a USB hub.

Talk to the build team

A row of Android phones on a rack used to host poker-bot sessions

My take in 30 seconds

Every "poker bot APK download" link on the open web is one of three things: malware that mines crypto, a fake login form for your room account, or a non-functional binary that opens once and crashes. I've tested dozens. None of them play poker. The few projects that play well never ship a public APK. They run on private custom builds, on real devices, behind a profiling agent, for clients who pay engineers to keep them alive.

If you want to play with a serious build, you talk to people. There's no Play Store listing. There's no Telegram drop. These builds are deliberately kept operator-side, not consumer-side. Public distribution kills them as soon as a rooms-side security team grabs a copy.

What an Android poker bot looks like on a rack

That's what a single bot "seat" looks like. The room's actual app runs on a real Android handset (the device fingerprint matters and rooms know how to read it). A separate process, the agent layer, reads the screen, decides actions, and posts taps back through input events. A third store holds opponent profiles, hand history, and per-seat economic state. On a real farm you have 20-200 of these seats stacked on shelves, USB-hubbed to a couple of mini-PCs, on residential or 4G IPs.

Three things that work

A

Real devices, not emulators

Bluestacks and Genymotion get flagged inside a session. Real handsets with fresh Google accounts, aged a week each, hold up much longer.

B

Per-opponent profiling

The agent has to soften vs fish and tighten vs regs at the same table. Static GTO leaks money fast in low-rake Asian apps.

C

IP discipline

One residential or 4G IP per device. Datacenter IPs are the fastest way to lose your bankroll inside a single afternoon.

Compatibility: tested devices

Device classReal-world stabilityFingerprintMy note
Pixel 4a / 5a (stock)GoodCleanStock firmware, no surprises.
Xiaomi Redmi 9-12 (MIUI)OkayMixedMIUI side-loads weird permissions, watch the agent.
Samsung A-seriesGoodCleanOneUI is fine, just disable Bixby.
Android emulators (Bluestacks etc.)BadBurntDetected. Don't.
Cloud phones (Redfinger etc.)BadBurntSame datacenter fingerprint, same outcome.

What the economics look like

A 60-seat farm built on Pixel 4a-class hardware costs me roughly $9-12K to stand up: devices, USB hubs, mini-PC controllers, decent power supply, racks, residential proxy contracts, and the agent licence itself. That's before any engineer time. Running cost per month sits around $1.8-2.4K in proxy and electricity. The bot loses ~20-25% of seats per month to bans, account flags, and dead handsets, and patching around that loss won't replace planning for it.

The other half is the agent layer. Even a strong agent eats half your edge between decision latency, the inability to see the table the way a human does, and rooms-side anti-collusion that downgrades winning accounts. Six-figure ROI happens in operator-grade clubs with controlled liquidity. It doesn't happen at NL10 on a public app, no matter what a Telegram seller tells you.

After six years: Device fingerprints matter. Each seat costs a lot to run, with the agent layer eating half your edge. Talk to us if you want a serious build.

Verdict

Android poker bots require an engineering project. Free browser downloads won't do the job.

6.4/10
my score, Android phone-farm category, June 2026