New ideaConsumer AI assistants

Mika-powered "phone agent" app

1 week ago · Scored 4 Nov 2026 · ID HBV-261104-02

Working title: Mika-powered "phone agent" app with no setup (managed model access, Play Store) for global and LatAm Android users — working name "Pocket Agent"

Summary

BriefingVerifiedHypothesisEstimate

Benchmark: wangzan101/Mika: 6,492 stars in 6 days, Apache-2.0, on-device agent that operates any app on a hidden background screen, learns skills from finished tasks; no Mika server (Verified, README and API). Existence check: No packaged/managed Mika app found. Analogs: Gemini agent features on Android, Meta and Chinese phone agents (Doubao, AutoGLM) — Hypothesis on feature overlap; open-source phone-agent repos exist.…

Read the rest

POS 52. On-device Android design and free phone agents from big tech are recorded as risks only, not cuts (Cesar, 4 Nov 09:36 BRT). Stars: Total 6,492 (github.com HTML 4 Nov ~12:00 BRT; API 6,492). Created 29 Oct 2026, Apache-2.0. 24h (3 Nov) +779; 7d +6,066 ≈867/day; since creation ≈1,082/day; 7v7 not computable (6 days); 3v3 3,336 vs 2,730 = +22%; peak 31 Oct (+1,480), 3 days since peak; 3 Nov −47% vs peak; today partial +456 (OSS Insight). Nothing estimated. X: 2,971 posts/7d (Falling -7% 3v3 (1–3 Nov 1,398 vs 29–31 Oct 1,505; daily full UTC days 9, 619, 877, 654, 556, 188); peak 31 Oct 877; 3 Nov is 79% below peak — ≥50% below peak: POP 5.3 window-closed signal). Revenue hypothesis: ~$1K–$5K/month at month 6 (HYPOTHESIS).

DescriptionHypothesisModel

A packaged version of the open-source Mika phone agent: install from the store, no API key, and say "order lunch with the best discount" or "summarise my group chat" while it works apps on a hidden background screen. Who: Android users (global, with Portuguese and Spanish first for LatAm) who want an assistant that actually taps through apps; prosumers who hate setup.…

Read the rest

Pain: Mika today needs a sideloaded APK and the user’s own model key or ChatGPT/xAI/OpenRouter sign-in; mainstream users will not do that. Monetization: Subscription with included task credits (model cost passed through) and a free tier of N tasks/month (Hypothesis; no approved price). Build: ~3–4 agent-weeks (fork, managed model proxy, onboarding, billing) plus Play policy review; stack: Mika fork (Kotlin, Apache-2.0), model proxy on Vercel/Fly, Stripe/Google Play billing, Pix. GTM: Faceless screen-recorded task demos on X/TikTok, SEO pages per task ("agente que pede comida com cupom"), referral credits.

Target customer

Android users (global, with Portuguese and Spanish first for LatAm) who want an assistant that actually taps through apps; prosumers who hate setup.

Pain

Mika today needs a sideloaded APK and the user’s own model key or ChatGPT/xAI/OpenRouter sign-in; mainstream users will not do that.

Why this score

Why it scored lower: The wave may be fading (X −79% from peak) while big-tech phone agents ship free, and Play Store policy on automation apps is unverified.

MoatHypothesis

Weak: Local-app skills library (iFood, Rappi, Mercado Livre) and Portuguese/Spanish tuning. Learned per-user skills make switching costly over time. Status: Hypothesis.

Demand signals

X posts (7 days)

2971

Signal trend

Falling -7% 3v3 (1–3 Nov 1,398 vs 29–31 Oct 1,505; daily full UTC days 9, 619, 877, 654, 556, 188); peak 31 Oct 877; 3 Nov is 79% below peak — ≥50% below peak: POP 5.3 window-closed signal

Web demandVerified

GitHub README (demos with WeChat, Amazon, Google Flights, food delivery); release APK downloads badge; iteration-1 theme count 3,249/7d (Verified, X API).

Market and competitors

TAM

Not sized this run — Scope: Global. Method: No third-party report fetched. (Basis: n/a)

SAMHypothesis

Not sized — Scope: Target markets above. Method: HBV assumption. (Basis: Hypothesis)

SOMHypothesis

~$12K–$60K ARR (Hypothesis) — Scope: Month-6 run-rate. Method: From revenue hypothesis ×12. (Basis: Hypothesis)

Competitors

Google Gemini on Android (agent features), Meta AI, ByteDance Doubao phone agent, Zhipu AutoGLM, open-source phone-agent repos; Mika itself free.

Business model

MonetizationHypothesisModel

Subscription with included task credits (model cost passed through) and a free tier of N tasks/month (Hypothesis; no approved price).

Revenue hypothesis (monthly)Hypothesis

~$1K–$5K/month at month 6 (HYPOTHESIS).

Jobs to be done

When I have small chores spread across five apps, I want to tell my phone once and let it do them in the background, so I can get my evening back.

HXC

Camila, 27, marketing analyst in São Paulo (illustrative) · Lives on WhatsApp, iFood and Mercado Livre; already pays for ChatGPT Plus. Buying signal: Uses ChatGPT daily but still taps through apps by hand.

Build and go-to-market

Tech stack

Mika fork (Kotlin, Apache-2.0), model proxy on Vercel/Fly, Stripe/Google Play billing, Pix

GTM channels

Faceless screen-recorded task demos on X/TikTok, SEO pages per task ("agente que pede comida com cupom"), referral credits

Risks

Main risk

The wave may be fading (X −79% from peak) while big-tech phone agents ship free, and Play Store policy on automation apps is unverified.

All risksVerifiedHypothesis

Value: Attention — X −79% from peak; daily stars −47% from peak (Verified) | Usability: Attention — Users fear an agent spending money | Feasibility: High — Play Store automation policy (Hypothesis) | Business Viability: High — Free big-tech agents

Next validation step

Re-read stars and X on 6 Nov (does decline continue?); read Google Play policy on accessibility/automation apps; check if Mika’s maintainer plans a store release.

Sources

github.com
github.com
OSS Insight
X API counts (window above)

newgithub-wrappergithubandroidagentsconsumerlatamglobal

Want the next one the day it is scored?

The score, the sources, the risks and the next step, by email.

ProductBots.ai

Like this idea? Turn it into a plan with ProductBots.ai.

ProductBots.ai is an Agentic Product OS: agents that do the product work with you, from discovery to plan. Download the briefing for this idea and use it as the starting point.

Open ProductBots.ai

Preview the briefing

# Briefing: Mika-powered "phone agent" app with no setup (managed model access, Play Store) for global and LatAm Android users — working name "Pocket Agent"

Source: Startup Ideas HQ (HB Ventures), found 2026-11-04. ProductBots Opportunity Score: 52/100, evidence confidence 50%.
Evidence labels (Verified, Self-reported, Hypothesis) are kept as written by the researcher.

## Goal for ProductBots.ai
Turn this idea into a validated MVP plan and a first build. Start from the next validation step; do not assume unlabelled numbers are verified.

## What it is
A packaged version of the open-source Mika phone agent: install from the store, no API key, and say "order lunch with the best discount" or "summarise my group chat" while it works apps on a hidden background screen.

Benchmark: wangzan101/Mika: 6,492 stars in 6 days, Apache-2.0, on-device agent that operates any app on a hidden background screen, learns skills from finished tasks; no Mika server (Verified, README and API). Existence check: No packaged/managed Mika app found. Analogs: Gemini agent features on Android, Meta and Chinese phone agents (Doubao, AutoGLM) — Hypothesis on feature overlap; open-source phone-agent repos exist. POS 52. On-device Android design and free phone agents from big tech are recorded as risks only, not cuts (Cesar, 4 Nov 09:36 BRT). Stars: Total 6,492 (github.com HTML 4 Nov ~12:00 BRT; API 6,492). Created 29 Oct 2026, Apache-2.0. 24h (3 Nov) +779; 7d +6,066 ≈867/day; since creation ≈1,082/day; 7v7 not computable (6 days); 3v3 3,336 vs 2,730 = +22%; peak 31 Oct (+1,480), 3 days since peak; 3 Nov −47% vs peak; today partial +456 (OSS Insight). Nothing estimated. X: 2,971 posts/7d (Falling -7% 3v3 (1–3 Nov 1,398 vs 29–31 Oct 1,505; daily full UTC days 9, 619, 877, 654, 556, 188); peak 31 Oct 877; 3 Nov is 79% below peak — ≥50% below peak: POP 5.3 window-closed signal). Revenue hypothesis: ~$1K–$5K/month at month 6 (HYPOTHESIS).

## Customer and pain
Target customer: Android users (global, with Portuguese and Spanish first for LatAm) who want an assistant that actually taps through apps; prosumers who hate setup.
Pain: Mika today needs a sideloaded APK and the user’s own model key or ChatGPT/xAI/OpenRouter sign-in; mainstream users will not do that.

## Market and competitors
TAM: Not sized this run — Scope: Global. Method: No third-party report fetched. (Basis: n/a)
SAM: Not sized — Scope: Target markets above. Method: HBV assumption. (Basis: Hypothesis)
SOM: ~$12K–$60K ARR (Hypothesis) — Scope: Month-6 run-rate. Method: From revenue hypothesis ×12. (Basis: Hypothesis)
Competitors: Google Gemini on Android (agent features), Meta AI, ByteDance Doubao phone agent, Zhipu AutoGLM, open-source phone-agent repos; Mika itself free.
Geography: Global; LatAm version in Portuguese/Spanish with local apps (iFood, Mercado Livre, Rappi); not China

## Business model
Monetization: Subscription with included task credits (model cost passed through) and a free tier of N tasks/month (Hypothesis; no approved price).
Revenue hypothesis (monthly): ~$1K–$5K/month at month 6 (HYPOTHESIS).

## Build and go-to-market
Build effort (agent-weeks): ~3–4 agent-weeks (fork, managed model proxy, onboarding, billing) plus Play policy review
Time to first revenue: ~8–10 weeks
Tech stack: Mika fork (Kotlin, Apache-2.0), model proxy on Vercel/Fly, Stripe/Google Play billing, Pix
GTM channels: Faceless screen-recorded task demos on X/TikTok, SEO pages per task ("agente que pede comida com cupom"), referral credits

## Risks and next step
Main risk: The wave may be fading (X −79% from peak) while big-tech phone agents ship free, and Play Store policy on automation apps is unverified.
Next validation action: Re-read stars and X on 6 Nov (does decline continue?); read Google Play policy on accessibility/automation apps; check if Mika’s maintainer plans a store release.

## Sources
https://github.com/wangzan101/Mika; https://github.com/wangzan101/Mika/releases; OSS Insight; X API counts (window above)

Research, not investment advice. Numbers keep the label they were published with.