Blog · Android

Quenderin for Android is in closed beta — how to get in

20 July 2026 · join the test

The native Android app is real and running. It’s not on the public Play Store yet — it’s in Google’s closed testing track, which has a two-step door. Here is exactly how to walk through it, and an honest account of what you get on the other side and what’s still rough.

Two steps, and why the first one is a Google Group

Google runs every closed test through an opt-in list. For Quenderin that list is a public Google Group, and joining it is literally what makes you a tester — there is no separate approval queue.

1. Join the testers group — groups.google.com/g/quenderin-testers. Use the same Google account that’s signed into the Play Store on your phone. This is the step people trip on: opt in with account A, try to install under account B, and Play will swear the app doesn’t exist. Membership is auto-approved — no waiting.
2. Opt in on Google Play — play.google.com/apps/testing/ai.quenderin.app. Tap Become a tester, then install. Prefer the store page directly? It’s here — but you still have to be in the group first. If Play 404s or says item not available, it simply hasn’t propagated to your account yet: give it a few minutes, occasionally an hour.

The “12 testers for 14 days” thing, said plainly

Google won’t let a closed test graduate to the public Store until at least 12 testers have opted in and stayed engaged for 14 continuous days. That’s the honest reason this reads like a recruitment drive: we need a dozen people actually opening the app across two weeks. Install-and-forget doesn’t count; a few opens spread over the fortnight does. If you join, that’s the whole favour — keep it on your phone and poke at it now and then.

What the Android build actually is

It’s the same local-inference core as the iPhone app: probe your hardware, recommend a model that fits, pull one GGUF file once, then run it fully offline through llama.cpp — rebuilt natively for Android, not wrapped in a web view.

The honest engineering note: iOS and Android don’t share a codebase. The core logic — the memory-fitness math that decides what your device can run, the model catalog, the download-integrity gate — is hand-ported between Swift and Kotlin. What keeps the two from drifting apart is a parity harness: a shared set of test vectors that both platforms must produce byte-identical answers for, checked on every commit in CI. It has caught real divergence between the twins more than once, which is exactly why it exists.

What’s rough, and what isn’t here yet

• It’s an early build. Expect UI rough edges, and on older or tighter-memory phones, model loads that are slower than you’d like. That’s the kind of thing this test exists to surface.
• A small local model is a small model everywhere. It will sometimes be confidently wrong. Every model carries a plain-language quality grade, same as on iOS — we’d rather tell you the ceiling than let you discover it mid-answer.
• The computer-use agent is macOS-only today. The part that operates a whole machine on your behalf lives in the Mac app. On Android, Quenderin is the private, offline chat — the model in your pocket, not desktop automation.

Tell us what breaks

The entire point of a closed test is the feedback. The group is the place for it, or file an issue on GitHub. If your device is on the older or lower-RAM end, that’s exactly the data we need — the memory probe is a heuristic, and real devices in real hands are how it gets less wrong. How that probe works, and where it can bite, is in the setup post.

Two links, that’s the whole ask: join the group, then opt in on Play. Thank you for helping it reach the finish line.