thymikee

Simlock leases, serves and manages iOS and Android devices for AI coding agents. You can turn your macOS and linux boxes into self-hosted cloud of simulators, emulators, and physical devices, that guarantees agents not fighting over the same device, control max number of simulators running at the same time to not clog CPU/RAM, run slim simulators with fewer services turned on, and many more.

We considered making it a paid but it makes more sense to just open it for everybody. If you have one or more spare Mac or Linux machines laying around, setup simlock and make it your private sim cloud for yourself or your team.

It's free, MIT licensed on GitHub: https://github.com/callstackincubator/simlock

Enjoy :)

Hi HN, I’m Michal. Together with Piotr, Artur, Lech, and Mike, for the last 8 months we've been building Apex, a coding model for mobile and web development. A model which we trained on React Native expertise and turned out to be good at React web too.

We're a small R&D team at Callstack, where we spend a lot of time working on React Native apps and create open-source libraries. We wanted to see how far that experience could take an open-weight model through domain-specific training. Could we achieve frontier-level performance for specific tasks at a much lower cost? After a few base model iterations, we sticked to Qwen as a foundation, trained on curated framework documentation, source code, APIs, and repository-grounded conversations, evaluated by our engineers.

On our React Native Evals (rn-evals.vercel.app), Apex scores 87.8%, compared with 88.6% for Claude Opus 5.5. The reported evaluation cost is $7.72 versus $28.75. These are our benchmarks, so we’d encourage you to try tasks from your own repository too.

Over the weekend we released a stealth version of Apex, Pixel Canary on Vercel AI Gateway. In Vercel’s Next.js evaluation, that preview passed 28 of 31 tasks, or 30 with documentation in context, matching GPT-6 Astra. Opening free access to the model on AI Gateway smashed our infrastructure and in these first days Pixel Canary was slow. We're not DevOps by train, but over these 5 days we had a quick crash course on scaling the model performance on B200s with custom prefill and decode setups to optimize load accordingly to the traffic we made, allowing us to achieve ~100 TPS at 4s p50 latency on the last day, with up to 80B tokens / day throughput.

The API is now available for business users (we need few more weeks to sort out EU consumer rights). It works with OpenAI- and Anthropic-compatible coding tools. We charge $0.50/M input, $0.20/M cache, and $3/M for output.

We’re interested in where specialization pays off and where it falls short. We're planning more specialized models for native iOS and Android development, and more.

I'm using this model for other languages and it's just as good as running Sonnet or Sol on medium. It's still Qwen, but tuning seems to make it better than barebones version on coding tasks.

If you try it, I’d love to hear which tasks it handles well and where you still reach for another model.

this will work with Flutter. agents will use screenshots instead of a11y snapshots instead. It will be just less efficient and a bit more token hungry

Callstack just announced a new framework for React Native. It's focused on reusable builds on your own cloud, supporting all available platforms and easy adoption in existing community CLI projects and native apps with some new brownfield packaging approach. For context: there's only Expo recognized as a framework in this space, so this creates some diversity. It's not yet open source, but it's gonna be in the upcoming weeks.

Callstack just announced a new framework for React Native. It's focused on reusable builds on your own cloud, supporting all available platforms and easy adoption in existing community CLI projects and native apps with some new brownfield packaging approach.

For context: there's only Expo recognized as a framework in this space, so this creates some diversity.

It's not yet open source, but it's gonna be in the upcoming weeks.

We just released a revised version of the guide from 2020.

Added 5 new chapters and 80+ pages. So now all it takes to write performant RN apps is just going through 200+ pages of tips, tricks, tools and tactics :D

Jokes aside. Optimizing mobile app perf without regressions is no easy task. It's no different with React Native. Some say it's less performant by default, some say the opposite. At Callstack we've helped dozens of teams optimize their React Native apps and as a result we came up with this free guide for anyone to grab.

Hope you'll like it! I was involved in creating this ebook and I'm happy to answer any questions you may have :)