Stringers
viajatech/agente-isabella · viajatech

A phone call where AI books a dentist appointment brings 170 stars to a Spanish-language GitHub repo

What it is A Spanish-language open-source Android app: after rooting your phone, it lets Gemini's real-time voice make actual phone calls over your own SIM and talk to the person who answers, no Twilio needed.

Why it took off The author posted a short video on TikTok and Instagram of AI calling a dentist for them, and the doctor's reaction got laughs. The TikTok had 226k views and the Reels had 64k likes, with a caption saying "free download on my GitHub." A landing-page-style README and an explicit ask for stars turned those non-developer viewers into stars and forks.

CHow you could follow this Don't copy the project itself. Copy the playbook: turn an open-source project you can demo into a short, native-language video set in everyday life, point the caption to GitHub, write the README like a landing page, and have an AI coding assistant walk people through installing it. Or port the same tech to other languages, like Portuguese or Indonesian.

C grade: the playbook is clear and worth learning, but almost all of its stars came from the author's own video, so the ceiling is around 180. Six or seven repos with the same tech never took off, and the root and Mac requirements make it hard to grow through word of mouth. Buzz in niche circles
Data (updated daily) · Repo created 2026-11-01 · 191 stars · 7 days +18
GitHub · stars
5010015020011-0711-1011-1311-1711-20191
Data 2026-11-07 → 2026-11-20 (14 days, 14 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

Mexico City tech creator David Ruiz built an Android app: after rooting, Gemini's live voice can place real phone calls from your own number. The same technique shows up in at least six other open-source GitHub projects, most with only 0 to 70 stars. This one picked up 178 stars and 33 forks in 13 days, and the driver wasn't the developer crowd. It was the author's own 88-second Spanish short video and a README written like a product landing page. Below: what the video showed, where the stars came from, who else is doing this, and how to try the same play for your own project's stars.

A call to the dentist

On the night of Nov 5, 2026 (UTC 21:22), @viajatech posted a video to TikTok. The screen is black, with two orbs that glow with the audio: a pink one labeled "IA DE DAVID" on the left and a blue one labeled "DENTISTA" on the right.

The call connects and the other end says: "Hello? Dental clinic, this is Dr. Ramírez." A female voice with a Mexican accent answers: "Good afternoon, Doctor, I'm Isabella, David's AI assistant, calling to help him book an appointment." The doctor pauses: "Well, after twenty years as a dentist, this is the first time a computer has called me." Isabella laughs: "Oh Doctor, 'computer' is so nineties. Just call me Isabella."

The doctor asks why David isn't calling himself. Isabella lowers her voice: "Between us? He's terrified of the drill..." In Spanish, "fresa" means both a dental drill and the Mexican slang for a pampered, soft person, and she makes a joke at David's expense. They settle on Thursday at five in the afternoon. She says she's already gone through David's calendar: "I manage his schedule, his emails, his reminders. The hard part is getting him there on time." Before hanging up, the doctor asks: "Could David build an assistant like yours for my clinic?" "Of course, I'll pass that along."

The video has no phone, no code, no install process, just github.com/viajatech/agente-isabella flashing at the bottom a few times. The caption reads: "Isabella can make domestic and international calls for you! Free download on my GitHub. Now the question... who would you have Isabella call?"

As of 11-14, the video has 226k views, 24k likes, 7,160 saves, and 2,170 shares on TikTok. The same video on Instagram Reels has 64k likes and 539 comments, more likes than on TikTok (Instagram, 11-14). The author has only 24k TikTok followers, and his usual videos get a thousand or two views. The last one to pass 100k was a Kali Linux video 2024 years ago.

The top comments are almost all off-topic: "Dr. Ramírez is in love" (2,656 likes), "Give her a Colombian accent and have her call him 'papasito'" (875 likes), "Just hearing her voice has me in love" (609 likes), "The movie Her is coming true." The technical questions come lower down: "The annoying part is the phone has to be rooted, does an emulator work?" (170 likes), "I was about to download it, then found out I need a Mac" (56 likes), "Is there an iOS version?"

What it actually is

A fully Spanish-language open-source Android app written in Java, built without Gradle. It places a call from your own SIM card. Once the other party picks up, they're talking to Gemini Live's real-time voice (voice name: Zephyr).

The hard part is that Android doesn't let ordinary apps inject audio into a call. The app gets around this by rooting the phone (with Magisk), then using the Qualcomm audio mixer to digitally feed the AI's voice into the VoLTE call, so it sounds clean with no echo. No Twilio, no rented numbers, no VoIP.

Several details are thoughtfully done:

  • The keys never live on the phone: the Gemini API key sits only in your own Cloudflare Worker (free), and the phone gets a one-time temporary token each time.
  • 8 safeguards on cost: every call has a cap, with per-minute and per-day limits. The call hangs up if the other side goes silent, and it won't reconnect after a drop. Before dialing, it asks for 3 seconds of confirmation to prevent accidental taps.
  • The first sentence must be "I am an AI", hardcoded in the settings and not editable from the phone.
  • You can use it without root: there's also a "no calls" mode where you chat with Isabella directly through the phone's mic and speaker.

The requirements are stated clearly: Android 14 or later, a rootable phone (rooting wipes the device, so use a spare), a Mac, a USB cable, your own Gemini key and Cloudflare account, plus a second phone to receive test calls. It has only been tested on a Samsung Galaxy A72.

Where the stars came from

The repo was created on 11-01 with 5 commits in total, and it hasn't been updated since 11-02 (GitHub, 11-14).

For the first four days, not a single fork. The author 11-01 had first posted a technical video titled "an autonomous agent that calls for you... no Twilio needed": 93 likes on Instagram, 2,453 views on TikTok, about 2,500 on YouTube Shorts. Same feature. Pitched as "no Twilio", nobody watched. Pitched as "it books my dentist and the doctor laughs", it blew up.

On 11-05 night the second video went out. About 7 hours later the first fork appeared (11-06 UTC 04:34), and 11-06 had 15 forks in a single day (GitHub). Star growth afterward:

DatestarAdded that day
11-0773—
11-08123+50
11-09136+13
11-10147+11
11-11154+7
11-12167+13
11-13173+6
11-14178+5

Many of the people who came weren't developers:

  • Signed up for GitHub just to "download": a spot check of 12 forkers found mostly default GitHub usernames in the "name+digits-hash" format, with 0 followers. One account was created 11-07 and forked the same day; another was created 11-10 and opened an issue the same day (GitHub).
  • Only two issues, neither a real problem: one is titled "Yahit," and the other pastes the README's git clone command verbatim as the issue title.
  • Forks are 18.5% of stars, far above normal for a project like this. Inference: many people assume a fork is just a "download."

Only one fork actually changed the code: pjuniot added automatic builds in 11-07, compiling the APK in the cloud and getting around the "you need a Mac" barrier.

Beyond these two videos, other channels are tiny. Search results show other people's AI and phone-related Instagram posts with comments like "Isabella, @viajatech built an Android app that makes real calls" and "comment ISABELLA and I'll send you the GitHub link," which look like the author driving traffic from other people's posts (only search snippets were seen, the original comments weren't opened to verify). Two directory sites that auto-index new GitHub projects picked it up, and a Japanese AI-generated podcast covered it in an episode. Nothing turned up on X, Hacker News, Reddit, or in the press. The author's last post on X is from July. It also never made GitHub's trending page.

Why this one turned views into stars

The same author's FLUX video from 2024 got 156k views, but its repo only picked up 2 stars. The difference is the repo itself: this README is written like a landing page.

  • At the top sit a self-made header image and four big buttons: "Watch the video," "Install," "How it works," and "It won't boot, what now?"
  • A tagline: "Your phone. Your SIM. Your calls. The voice is Isabella's." Below it, a call-bubble diagram and a nine-step install flow chart, all in Spanish.
  • The install instructions aren't a list of commands. They say: "Clone it, open your AI coding assistant, and tell it: read AGENTS.md and walk me through installing Isabella on my phone step by step." The repo ships AGENTS.md, CLAUDE.md, and GEMINI.md, so Claude Code, Codex, Gemini CLI, and Cursor can all use it. According to the README, the AI handles the computer-side work and stops to wait for you on things like pressing volume keys or confirming a wipe.
  • There's a dedicated "unbricking" doc, recording how they recovered their own Galaxy A72 after it got stuck in recovery mode post-root.
  • The closing line is direct: "If you like it, give the repo a star so more people can see it."

Timing matters too: Google 9-24 started testing Gemini "Call for Me," an AI that phones businesses for you. TechCrunch and Wired covered it, but it's only for US Pixel 11 users. Spanish speakers have heard of the concept but can't use it. Isabella claims it works on any rooted Android with your own number, for domestic and international calls, which fills exactly that gap.

The repo isn't the end goal for the author. His site viajatech.com sells website building (from 1,500 Mexican pesos, ordered via WhatsApp) and AI services for business owners, and another site sells ebooks and courses. The sales rep on his site is also named Isabella. The doctor's line at the end of the video, "could you make one for my clinic too," is itself an ad. Stars are a side benefit for his personal brand.

Who else is building

The "rooted Android + your own SIM + real-time voice AI" idea isn't new:

ProjectLaunchedScale (November)How it differs
Google Gemini "Call for Me"Testing since 2026-09-24US Pixel 11 onlyOfficial, no root, but only calls businesses and only in the US
Bland, Vapi, Retell, Genspark Call For Me, etc.2023–2025Commercial platformsRun on Twilio or rented numbers, billed per minute
sidinsearch/AgentCall2026-07-3068 starNearly the same mechanism, in English, aimed at agent developers, with no videos
mutonby/agentring2026-10-23420 starRun over VoIP; the AI calls you to report back, but doesn't dial out for you
pipecat-ai/pipecat, dograh-hq/dograh2023-12 / 2025-0917k / 6,100 starsGeneral-purpose voice agent frameworks, connected to phone lines via Twilio / SIP
vendoriaai/GeminiPhoneAI2025-10-228 starAlso uses Gemini Live, mostly for answering and screening incoming calls
CallGateAI, callrelay, sim-phone-mcp, bageo-call2026-03 to 110–1 starAll "Android SIM + voice AI," none promoted

The clearest comparison is AgentCall: it came out three months earlier, does nearly the same thing, is in English, and has 68 stars. What Isabella added is an author who makes short videos, a Spanish-speaking market, and a README that feels finished. The "no Twilio" angle gave the video a sellable point, but it went viral because the call was funny and the voice sounded human.

What it doesn't do

  • The bar is too high for most people: it needs root, a Mac, and has only been tested on one phone. These three issues are the most common questions in the comments. How many people actually got it working is unknown; inference: very few.
  • No ready-made APK: you have to build it yourself. pjuniot's fork already builds in the cloud, but it hasn't been merged upstream.
  • Spanish only: the README, persona, and install flow are all in Spanish.
  • Stalled: no commits since 11-02, daily stars fell from 50 to single digits, and the author hasn't posted a follow-up video.
  • No iOS: iPhones can't inject audio into calls, and nobody can fix that.

If it were you

Verdict: C. The playbook is worth learning, but following this project directly isn't worth the effort. Nearly all of its stars came from the author's own video, and the ceiling sits around 180. Six or seven other repos already do the same thing and none of them took off. The root and Mac requirements make it hard for word of mouth to carry it further. What you can take away is that short-video viewers will star a GitHub repo: an account with 24k followers that usually pulls a thousand or two views is enough for one well-timed video.

Who it fits: developers who already have an open-source project that can be shown off (AI calling, AI replying to messages, AI operating a phone, where the result is obvious at a glance), who are willing to make short videos in their native language, and who have a few thousand to tens of thousands of followers on TikTok, Reels, or Douyin.

Who it doesn't fit: people who won't appear on camera or do voiceover and want to rely only on organic GitHub traffic (AgentCall is the 68-star example); anyone wanting to build another "AI phone calls" product for Spanish speakers, since that spot is taken; and anyone whose project only developers can understand and that can't be turned into an everyday scene.

How to do it, ranked by how worthwhile:

  1. Applying the playbook to your own project: pick an everyday scenario (booking a dentist, chasing a delivery, haggling with a landlord), record a real or staged conversation under 90 seconds. A black background with a voice visualizer is enough. Don't explain the mechanism; show the other person laughing. Caption it "Free, download it from my GitHub," with the link on screen. At the same time, rewrite your README as a landing page: native language, a header image, big buttons, diagrams, and "have your AI coding assistant read AGENTS.md and install it for you." End by asking for a star, plainly.
  2. Switch languages and do the same thing: Portuguese (Brazil), Indonesian, Arabic. These are markets Google's "Call for Me" won't cover anytime soon. Same tech, plus a local-language video. You can start from an open-source project in the same space, but check each license first.
  3. Fill what it's missing: build a version that doesn't need a Mac, ships prebuilt APKs, and supports more phone models. Or pull its root-free speakerphone mode out as its own app you can install directly. This earns reputation. Stars will come in lower numbers than the first two options, and confirm the license allows it first.

How to get users: for the first version, film everyday-life scenarios, not tech demos. Its tech version got 93 likes, the scenario version 64k likes. Post to TikTok and Instagram Reels at the same time; it did much better on Reels. Ask a question in the comments ("Who would you have it call?") to drive engagement. Leave comments mentioning your project under popular posts on the same topic. Put the video link in the README and the repo link in the video, so each points to the other.

Timing: before Google rolls out "Call for Me" beyond the US and to more Android models. After that, "AI makes calls for you" won't be new, and Spanish and Portuguese speakers will have an official root-free option.

Watch out: AI calls to real phone numbers are regulated in many places. The US FCC already put AI-synthesized voice autodialing under the TCPA in 2024. Demo it by calling people you know, and don't let anyone use it for mass calling.

Signs you've read this wrong: ① The author stops posting videos, but the repo keeps adding 10+ stars a day steadily through late November. That would mean GitHub traffic on its own is bigger than estimated, and a complete version is still worth building. ② Someone follows this playbook, makes a video that gets 100k views, and gets only a few dozen stars. That would mean this one relied on luck in the Spanish-speaking market and on this README, not on a repeatable method. ③ The reverse: if short videos keep bringing projects that all gain real traction, this channel deserves more serious effort than a C grade.

Data and sources

GitHub: viajatech/agente-isabella repo page, README, commit history, fork list, issues (11-14); daily stars (11-07 to 11-14); star counts of similar repos (November). TikTok: caption, plays and comments for video 7704800773791599880 (11-14); the video is a retelling of the original clip, dialogue translated from Spanish. Instagram: likes and comment counts for Reels DfX9OGvh4kt and DfN7L7vuWbt (11-14). YouTube Shorts: oSUbPbwd6TY. Google "Call for Me": TechCrunch, Wired 2026-09-24. Author business: viajatech.com. Anything marked "inferred" or "unverified" is inferred or not yet checked.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report
Competitors
NameWhereLaunchedScaleHow it differs from this one
Google Gemini "Call for Me" (and Duplex / Ask for Me)Big-company featuresTesting since 2026-09-24US Pixel only 11Official, no root needed, but only calls businesses and only in the US. Isabella can call any number, works in Spanish, requires root.
Genspark Call For Me / Bland / Vapi / RetellOther2023–2025Commercial SaaS/APIRuns on Twilio or rented numbers, billed per minute. Isabella's pitch is using your own SIM and your own API key.
sidinsearch/AgentCallOpen source2026-07-3068 star / 19 forkNearly the same mechanism (rooted Android as a cellular call gateway), but in English and aimed at agent developers. No content-driven reach.
mutonby/agentringOpen source2026-10-23420 star / 67 forkRuns over VoIP. The AI calls you to report back. It doesn't make outbound calls on your behalf.
dograh-hq/dograh, pipecat-ai/pipecatOpen source2025-09 / 2023-126.1K / 17.3K starGeneral voice agent frameworks that rely on Twilio/SIP. Isabella is a single finished product that connects directly from the device.
vendoriaai/GeminiPhoneAIOpen source2025-10-228 starAlso uses Gemini Live, but focused on answering and screening incoming calls. No packaging or promotion.
CallGateAI / callrelay / sim-phone-mcp / bageo-callOpen source2026-03 to 110–1 starAlso "Android SIM + real-time voice AI". Nobody promoted them.
chenxiaolong/BCPOpen source2022-06-2218 starA technical demo that injects audio into calls. A predecessor of the underlying idea.
Where users come from
  • Verified The author's own TikTok: video posted 11-05 21:22 UTC with 7704800773791599880 views, 226,307 likes, 24,338 favorites and 7,161 saves. Caption: "Descarga a isabella gratis en mi github" (TikHub).
  • Verified Author's own Instagram Reels: DfX9OGvh4kt (11-05): 63,949 likes, 539 comments (IG embed page)
  • Verified Launch technical video (didn't take off): 11-01's "Agentes Autónomos…sin Twilio": 93 likes on IG, 2,453 plays on TikTok, 2.5K plays on YouTube Shorts
  • Inferred Commenting on other people's posts / comment-keyword DMs: Jina search snippets show comments under IG posts by embarca.vc, bancolafisepa and iphoneveneno: "Isabella, de @viajatech, es una app Android…" and "Comenta ISABELLA y te mando el GitHub". The original comments weren't opened to check.
  • Verified README as landing page + star requests: custom SVG header and buttons, AI coding assistant install instructions (AGENTS/CLAUDE/GEMINI.md), closing with "regálale una estrella". Some forkers are GitHub accounts created that same day.
  • Verified Aggregator sites / auto-generated podcast: listed on SkillsLLM and GitReverse. Spotify's "Kana & Mari's SoundRepos" covered it in an episode.

Why viajatech/agente-isabella (Agente Isabella) took off

Bottom line

It wasn't discovered by the developer crowd. David Ruiz, a Mexico City tech short-form creator (@viajatech, 24k followers on TikTok, 32k on Instagram, 22.5k on X), filmed a Spanish-language video of an AI calling his dentist on his phone number to book an appointment. The AI-calling-the-doctor part is what took off (the top comment is "Dr. Ramírez fell in love"). The caption said "free download on my GitHub," and from Nov 5 on, a wave of newbies starred and forked the repo. The repo itself reads like a landing page: a homemade SVG header, buttons, and "have your AI coding assistant read AGENTS.md and walk you through the install," ending with a direct request for stars. The stars are basically a side effect of his personal brand and his freelance business (website building, AI services).

Key facts

  • The repo was created 2026-11-01 05:11 UTC, has 5 commits, and the last push was 11-02. No updates since (GitHub API).
  • Stars / forks (our github repo_day): 11-07 was 73/12,11-08 and 123/22 (+50 in a single day at peak). 11-09 was 136/26,11-14 and 178/33. After that, +5 to 13 per day, declining. It never made trending, and a repo under 60 days old can't get on it.
  • fork/star ratio about 18.5%. Fork timing (GitHub forks API): the first fork landed on 11-06 04:34 UTC, 11-06 had 15 forks all day, 11-07 had 8, then only sporadic ones. None before that, so for about 4 days after 11-01 launched, nobody knew it existed.
  • Only 2 issues, none of them technical: #1 titled "Yahit"; another (11-14) pasted the README's git clone … command verbatim as the issue title. So a large share of the people arriving were non-developers.
  • Spot-checked 12 fork accounts: antonifuentes9866-hash was created 11-07 and forked the same day; isabelvazquez5821-alt was created 11-10 and opened an issue the same day; alvarezdlcjose-rgb was created 11-03. Most follow the default GitHub newbie username pattern of name + number-hash/-crypto/-alt, with 0 followers (GitHub users API). In other words, some people watched the video and went and signed up for GitHub specifically to "download" it.
  • Only pjuniot's fork made substantive changes: 11-07 added CI and builds the APK in the cloud, getting around the "you need a Mac" barrier.

Where users come from (growth levers)

  1. Author's own TikTok hit (confirmed): posted 2026-11-05 21:22 UTC, "Isabella puede hacer llamadas telefónicas nacionales e internacionales por ti! Descarga a isabella gratis en mi github" (aweme 7704800773791599880): 226,307 plays, 24,338 likes, 2,169 shares, 7,161 saves, 209 comments (TikHub, scraped 11-14). His account's usual videos get only 1–2k plays. This is his biggest post since the Kali Linux video (2024 years ago, 1.03M plays) and FLUX (156k). The timing fits: the first fork came about 7 hours after release.
  2. Same video on Instagram Reels (confirmed): DfX9OGvh4kt, posted 11-05, 63,949 likes and 539 comments (Instagram embed page). Going by likes, IG may have driven even more reach than TikTok.
  3. The launch video flopped (confirmed): 11-01 first posted a more technical one, "Agentes Autónomos que marcan por ti…sin necesidad de usar Twilio". It got only 93 likes on IG (DfN7L7vuWbt), 2,453 plays on TikTok, and 2.5K plays on YouTube Shorts. 11-05 The second version switched to everyday scenarios: booking a dentist appointment, the doctor getting flustered by the AI, and ending with "Who would you let Isabella call?" That version took off. The YouTube version (tAx8WNWMqc8) got only 1.3K plays. YouTube was not a channel that worked.
  4. Comment seeding (partly unverified): search results show third-party IG posts (embarca.vc, bancolafisepa, iphoneveneno and others about AI and phone calls) with comments like "No es tan ciencia ficción: Isabella, de @viajatech, es una app Android que hace llamadas reales…" and "Comenta ISABELLA y te mando el GitHub" (Jina search snippets). This looks like the author or accounts tied to him commenting on other people's posts to drive traffic, plus a "comment a keyword and get the link by DM" pitch. The original comments weren't opened to verify.
  5. Aggregators and machine repackaging (confirmed, small reach): SkillsLLM and GitReverse list it. Japan's AI-generated podcast "Kana & Mari's SoundRepos" did an episode titled "AIが電話するAndroidアプリ" (Spotify). These pick up new GitHub repos automatically and don't bring much traffic.
  6. Channels that didn't show up: x-radar searches for "viajatech" and "agente-isabella" over the last 30 days returned 0 results each. The author's last post on X was in July. No coverage found on HN, Reddit or in the media.

Why now, and why people click through

  • Hot topic: Google 2026-09-24 is starting to test Gemini "Call for Me," which lets AI phone businesses for you (TechCrunch and Wired both covered it), but it's only for Pixel 11 users in the US. Spanish-speaking users can see the concept but can't use it. Isabella claims "any rootable Android, your own SIM, domestic and international calls," which fills exactly that gap (Wired / TechCrunch 2026-09-24).
  • Content hooks are entertainment, not tech: the top TikTok comments are "El dr ramirez se enamoro" (2,656 likes), "add a Colombian accent so she calls you papasito" (875 likes), "just hearing her voice and I'm in love" (609 likes), and "the movie *Her* is becoming real." The people asking technical questions concentrate on "do I need root" (170 likes), "do I need a Mac?" (56 likes), and "is there an iOS version?" The people starring the repo are mostly won over by Gemini Live's Zephyr voice and realistic conversation, not by any real plan to deploy it.
  • The repo was turned into a landing page non-developers can click through: fully in Spanish, with a custom SVG header, four big buttons ("Watch video / Install / How it works / Unbrick"), a sample call-bubble illustration, and a staged flow diagram. The install instructions say "clone it, open your AI coding assistant, and have it read AGENTS.md to walk you through step by step" (CLAUDE.md and GEMINI.md are included too). It ends with "⭐ Si te gustó, regálale una estrella al repo: así le llega a más gente." Compare his FLUX video from 2024 years ago, which got 156k views, with the 2 stars on the matching repo. Same kind of traffic, much higher conversion this time thanks to the repo packaging.

Competitors

NameWhereLaunchedScaleHow it differs from Isabella
Google Gemini "Call for Me" / the earlier Duplex, Ask for MeBig-company featuresTesting since 2026-09-24US Pixel only 11Official and root-free, but it only calls businesses and is limited to the US. Isabella dials any number, speaks Spanish, and requires root.
Genspark Call For Me, Bland, Vapi, Retell, SynthflowOthers (SaaS / API)2023–2025Commercial platformsThey go through Twilio or rented numbers and charge per minute. Isabella's pitch is precisely "no Twilio, use your own SIM."
sidinsearch/AgentCallOpen source2026-07-3068 star / 19 forkNearly identical mechanism (rooted Android as a cellular call gateway), English, aimed at agent developers like Hermes/OpenClaw users, with no content driving reach
mutonby/agentringOpen source2026-10-23420 star / 67 forkGoes over VoIP. The AI calls *you* to report back, rather than calling out on your behalf. English developer circles.
dograh-hq/dograh, pipecat-ai/pipecatOpen source2025-09 / 2023-126.1K / 17.3K starA general voice-agent framework that reaches phones via Twilio/SIP. Isabella is a single finished product that connects directly from the device.
vendoriaai/GeminiPhoneAIOpen source2025-10-228 starAlso uses Gemini Live on Android to take calls, but leans toward answering and screening incoming calls
h4lo-void/bageo-call (Bagei)Open source2026-11-121 starRooted Pixel + Gemini Live answering real numbers, same approach, same period. The README doesn't mention Isabella, so it isn't a copy.
compwiz21/CallGateAI, highsierraloft/callrelay, nwcnwc/sim-phone-mcpOpen source2026-03 to 100 starAll are "Android SIM + real-time voice AI" gateways. None of them got promoted.
chenxiaolong/BCPOpen source2022-0618 starA technical demo that injects audio into calls. A predecessor of the underlying idea.

Bottom line: the technique of "rooted Android + SIM + real-time voice AI making phone calls" isn't new. There are at least 6 comparable open-source projects, most with 0–70 stars. What Isabella adds is an author who makes short videos and has 50k followers, plus a Spanish-speaking market and a README that reads like a product page.

Differentiation

  • Technically: it injects audio digitally into VoLTE calls using the Qualcomm mixer (tinyplay/tinycap), so there's no echo. A Cloudflare Worker issues only one-time Gemini tokens, so the API key never lives on the phone. 8 cost safeguards (caps per call, per minute, and per day; auto-hangup on silence; no reconnects). The first sentence is forced to declare "I am an artificial intelligence." It skips Gradle (aapt2→javac→d8→apksigner). There's also a root-free "phone speaker conversation" mode. Only tested on a Galaxy A72.
  • Compared with rivals, what really sets it apart is: Spanish, a finished-product feel (packaging, unbricking docs, an AI assistant that helps you install), and the author's built-in audience.
  • Did the differentiation make it take off? Only half. Nearly all the stars trace back to the author's own short video. The technical differentiator (no Twilio, your own SIM) gave the video a sellable hook, but the video took off because the calls were funny and the voice was convincing. The repo packaging decided how much of that traffic turned into stars. The same tech in an English developer repo like AgentCall only got 68 stars.

How many people know about it

It's circulating in niche circles. It got hundreds of thousands of views on Spanish-language TikTok and IG, but almost nobody in the GitHub ecosystem mentioned it: it didn't hit trending, and there's no coverage on X, HN, Reddit, or media, only auto-aggregator sites picking it up. It's open source and free, with a license that prohibits commercial use, so no revenue is involved.

Copycats

I didn't find a direct copy. The only thing is a pjuniot fork that added a CI build for the APK. The contemporaneous bageo-call (11-12, rooted Pixel + Gemini Live answering calls) takes the same approach, but there's no sign it was inspired by this one.

Where the money comes back (the author's books)

The repo is the funnel for his personal brand: viajatech.com sells website building (from 1,500 Mexican pesos, ordered via WhatsApp), there's an "IA para Empresarios" AI service, and viajatech.xyz sells e-books and courses (Shodan OSINT, a manual on nootropics, and more). On the official site, Isabella is the web voice/text salesperson (viajatech.com/isabella), and his books are also listed on Amazon. In April 2026 he previewed this persona on X with a trilingual post (English, Spanish, Japanese): "Soy Isabella, la primera agente autónoma…" Getting stars doesn't earn money directly. What it earns him is name recognition and credibility on GitHub.

What to borrow

  1. Rework the tech demo into a life scenario: the same feature as "an autonomous agent without Twilio" got 93 likes, while "AI calls my dentist for me, and the doctor gets flustered flirting" got 64k likes.
  2. The short-video copy says outright "free, download it from my GitHub," treating GitHub like an App Store, so non-developers sign up for accounts and click star or fork.
  3. Write the README like a landing page: native language, big buttons, illustrations, unbricking docs, "have your AI coding assistant walk you through installation" (AGENTS.md / CLAUDE.md / GEMINI.md all included), and an explicit request for a star at the end.
  4. Beat the big companies by shipping features they release in only one country, on one device model (Gemini Call for Me is only for US Pixel 11), and use open source plus local language to claim other markets first.
  5. Use an open-source repo as the storefront for a personal freelance business, and use the same persona (Isabella) to tie together the short videos, the website salesperson, and the repo.

Unknowns

  • The timestamps for stargazers aren't available (API 401/403), so matching stars to the video timing can only be inferred indirectly from the fork times and our daily snapshots starting from 11-07.
  • IG reel view counts aren't accessible, only likes and comments.
  • For "leaving comments under others' posts to drive traffic," I only saw search summaries and didn't open the original comments to verify.
  • The number of people who actually got it deployed is unknown. It requires root and a Mac, and was only tested on an A72, so probably very few. The only 2 issues also suggest most people who came couldn't get it running.

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.