ML & Android @ Droidrun https://xitter.gq/t.co/JPEB4YXrE5
40°24'53.6"N 49°50'44.3"E
Joined January 2011
- Tweets152
- Following63
- Followers21
- Likes284
Pinned Tweet
mobile-harness achieved a 98.34% mean score on the iOSWorld benchmark!
mobile-harness gives agents like Codex, Claude Code, and OpenClaw better mobile controls, app cards, and persistent memory. Agents can learn from previous runs by reviewing what worked and what didn’t, then updating their cards and memory for future tasks.
For iOSWorld, we used codex with gpt-6-astra / medium and gpt-6-luna / high.
Across 133 tasks, it received 123 full scores and 10 partial scores. Its binary pass rate, which counts only full scores, was 92.48%.
Explore the results and task replays: mobilerun.ai/benchmark/
@droid_run @mobilerun_ai
Rasul Osmanbayli retweeted
🚀Mobilerun just crossed 10k GitHub stars.
Huge thank you to all of you for the support since day one! 🙏
Here are 5 repos for builders who want to automate real phones:
Rasul Osmanbayli retweeted
The next generation of companies will be built and run by teams of AI agents.
We’re building the mobile infrastructure they’ll need. Excited to see what starts this weekend.
We're joining @ycombinator × @usenaive as the Mobile Side Track at the world's first Autonomous Company Hackathon.
Build with the Naïve SDK and put your agents
to work in real mobile apps.
>72h of Android + iOS minutes on Mobilerun.
> iPhones are first come, first served.
> Winners take home annual cloud iPhone packages.
Nov 7–8 · YC office, SF
Pick a business. Ship the agent team that runs it.
Demo it live.
Sponsors: @OpenAI @OpenRouter @supabase @neondatabase @vercel @stripe @agentmail @tenki @MonidHQ @apify
Further details ⬇️
Rasul Osmanbayli retweeted
An autonomous company will eventually run into an app with no API and no useful website. I want to see what people build when that stops being a blocker.
We're giving teams real cloud iPhones and Androids at the Naïve hackathon at YC this weekend.
Curious to see what people build that we haven't thought of ourselves.
We're joining @ycombinator × @usenaive as the Mobile Side Track at the world's first Autonomous Company Hackathon.
Build with the Naïve SDK and put your agents
to work in real mobile apps.
>72h of Android + iOS minutes on Mobilerun.
> iPhones are first come, first served.
> Winners take home annual cloud iPhone packages.
Nov 7–8 · YC office, SF
Pick a business. Ship the agent team that runs it.
Demo it live.
Sponsors: @OpenAI @OpenRouter @supabase @neondatabase @vercel @stripe @agentmail @tenki @MonidHQ @apify
Further details ⬇️
Rasul Osmanbayli retweeted
We are bringing mobile automation to the Naïve hackathon at Y Combinator.
Build agents that interact with real Android and iOS apps.
Join us for the Naive hackathon this weekend at Y Combinator, SF.
Register here 👇
We're joining @ycombinator × @usenaive as the Mobile Side Track at the world's first Autonomous Company Hackathon.
Build with the Naïve SDK and put your agents
to work in real mobile apps.
>72h of Android + iOS minutes on Mobilerun.
> iPhones are first come, first served.
> Winners take home annual cloud iPhone packages.
Nov 7–8 · YC office, SF
Pick a business. Ship the agent team that runs it.
Demo it live.
Sponsors: @OpenAI @OpenRouter @supabase @neondatabase @vercel @stripe @agentmail @tenki @MonidHQ @apify
Further details ⬇️
Rasul Osmanbayli retweeted
We're joining @ycombinator × @usenaive as the Mobile Side Track at the world's first Autonomous Company Hackathon.
Build with the Naïve SDK and put your agents
to work in real mobile apps.
>72h of Android + iOS minutes on Mobilerun.
> iPhones are first come, first served.
> Winners take home annual cloud iPhone packages.
Nov 7–8 · YC office, SF
Pick a business. Ship the agent team that runs it.
Demo it live.
Sponsors: @OpenAI @OpenRouter @supabase @neondatabase @vercel @stripe @agentmail @tenki @MonidHQ @apify
Further details ⬇️
Rasul Osmanbayli retweeted
Under-discussed problem with mobile automation: it doesn't scale past one phone
Now you can run workflows across multiple iPhones at the same time
Or plug in your own iPhone and your agent can control any app on it without any emulator.
Rasul Osmanbayli retweeted
Setting up an an ai agent on your iphone takes less time than making coffee ☕️
1. Plug it in
2. Pair it in your browser
3. Type a prompt
That's it. No app on the phone, no Xcode, no jailbreak.
Here it's checking my LinkedIn connection requests and summarizing the important ones.
Try it right now, no account needed 👇
Rasul Osmanbayli retweeted
We just hit 98.34% on iOSWorld Benchmark.
We ran the full 133-task iOSWorld benchmark and mobilerun achieved:
> 123 full scores + 10 partial scores
> 98.34% mean official score
iOSWorld tests AI agents across complex, multi-step workflows on iOS apps.
You can inspect every task, trajectory, action, and replay yourself👇
Rasul Osmanbayli retweeted
YOUR AI AGENT CAN NOW CONTROL A REAL IPHONE.🔥
Just connect your iPhone to Mobilerun Cloud and let an agent control it with a single prompt.
Super simple setup.
Just an iPhone, a Mac or Linux, and under 2 minutes to get started.
Yes, you can even use Claude or Codex to run the automation😉.
Rasul Osmanbayli retweeted
Huge shoutout to the tech team @megauega, @nielsmdt99, Lucius, @Rasul_ for building such a great product. Proud of what we're creating together.
Rasul Osmanbayli retweeted
Copycats are popping up left and right across every industry.
Competition is great. It pushes the market forward.
Still waiting to see something that truly rivals @mobilerun_ai .
Rasul Osmanbayli retweeted
Most phone agents are tested in simulated environments, but the real world is messier: carts persist, listings change hourly, and navigation varies.
At @ScaleAILabs, we ran agents on real phones with real apps on AgentEnv Framework.
See more at agentenvframework.com/blog/a…
Rasul Osmanbayli retweeted
@elonmusk when can we provide you with the mobile runtime with our phone infra?