#callstack
AI agents can now drive apps

A TypeScript library that lets AI agents control mobile, web, TV and desktop apps.

🍒 Built by the Callstack team
github.com/callstack/agent-device

#opensource #typescript
November 18, 2026 at 8:53 AM
some of my takeaways from this talk by Tejas Kumar.
- AI != all humans are software devs. We have great cameras on our phones, and wedding photographers still have plenty of work
- using MCP, you can also pass UI templates to display the fetched data

youtu.be/aNZz40sQg_g
The Critical Advantage With AI | Tejas Kumar at Agent Conf 2026
YouTube video by Callstack
youtu.be
November 17, 2026 at 9:18 AM
🚨 EUVD-2026-125549
📊 7.1/10
🏢 callstack

📝 agent-device from 0.21.36 before 0.21.37 contains a local file disclosure vulnerability that allows authenticated remote callers to read daemon-host f...

🔗 https://euvd.enisa.europa.eu/vulnerability/EUVD-2026-125549

#cybersecurity #infosec #cve #euvd
November 13, 2026 at 5:03 PM
i have a crash report in production, and thanks to @sentry.io i have a callstack and a minidump, but unfortunately i couldn't make sense of it. my cousin claude however was pretty good at figuring it out. if you played terralysia today and it crashed, your cpu might be busted
November 11, 2026 at 1:30 AM
A new AI review! callstack/agent-device ⭐3.9/5.0
agent-device is a substantial, well-maintained TypeScript monorepo that provides a device-control and evidence-capture layer for AI coding agents across iOS/Android (simulators, emulators, and physical de...
https://gitrated.com/callstack/agent-device
November 5, 2026 at 1:53 PM
AI is changing frontend fast. Ask the people behind the tools live next Friday, Nov 13. Free.

AI.js Day 2026 broadcast with Danny Thompson, Marina Wyss, Kent C. Dodds, and experts from Figma, Chrome, Storybook, MUI, Callstack, and JetBrains.
Save your spot: https://jb.gg/nr49ip

#AIJSDay2026
November 3, 2026 at 11:23 AM
Storybook Connect brings Storybook into WebStorm and other JetBrains IDEs. Includes a CLI and skills for you and your agent.

Built with Callstack and Storybook teams.
Install it and tell us what you think: https://jb.gg/lo5687
November 2, 2026 at 9:52 AM
Agentic workflows require dedicated UI primitives. shadcn released components built specifically for agent traces and code execution.

To test these autonomous agents, Callstack introduced Appduct, providing programmatic hooks into React Native state to verify complex UI pipelines.
October 25, 2026 at 3:45 PM
React Native AI Just Updated: Version 0120 Brings AI SDK v6 Compatibility, Llama Tool Calling, and Re-Ranking + Video

A Major Step Toward More Capable On-Device AI The React Native AI project, maintained by Callstack Incubator, has released version 0.12.0, introducing a set of improvements…
React Native AI Just Updated: Version 0120 Brings AI SDK v6 Compatibility, Llama Tool Calling, and Re-Ranking + Video
A Major Step Toward More Capable On-Device AI The React Native AI project, maintained by Callstack Incubator, has released version 0.12.0, introducing a set of improvements designed to make AI integration in React Native applications more capable and compatible with the latest AI SDK architecture. The update brings all providers in line with the V3 provider specification for AI SDK v6, adds tool-calling support for Llama models, and introduces re-ranking capabilities for Llama-based workflows.
undercodenews.com
October 24, 2026 at 10:42 PM
Test Automation with agent-device (Part 1)
agent-device is an open-source tool from Callstack that lets an AI agent drive iOS and Android apps and check what is on screen. This post covers why it was picked for testing a Flutter app, how to install it for Claude Code, the basic command loop, and what to watch out for with Flutter. It is based on version 0.21.31 (2026-10-24). ## Why agent-device Flutter ships Android and iOS apps from one codebase, so the device-testing tool has to cover iOS too. Google Artemis, the first candidate, is a natural-language testing agent, but it supports Android only; iOS is still just a roadmap item. The other candidates looked like this: * **mobile-mcp** exposes taps, screenshots, and the accessibility tree on iOS and Android over MCP (the protocol an AI uses to call external tools). It has no sessions, replay, or evidence storage, so a verified flow is hard to hand off to CI. * **Maestro** is an E2E framework that pins flows in YAML, and it has an MCP server. It suits regression runs of fixed flows better than an agent exploring the screen. * **Arbigent** is an AI agent built on top of Maestro. Its author rates it 1/5 for speed and 3/5 for reliability, and its YAML format was still changing. * **Commercial services** (Drizz, Minitap, and others) bundle a real-device farm with a natural-language agent, but the only public material was vendor material, so cost and quality could not be checked. agent-device handles iOS and Android simulators, emulators, and real devices with one tool, and its docs state Flutter support explicitly. It ships both a CLI and an MCP server, can save a verified flow as an `.ad` script to replay later, and can export to Maestro YAML. It is MIT-licensed. ## Installation Install the CLI globally and check the environment. npm i -g agent-device@latest agent-device doctor `doctor` checks connected devices, toolchains, and dev-server status. In this setup it passed with only warnings for unused platforms (HarmonyOS, Vega). Register it with Claude Code as an MCP server. With `-s user`, the same tool is available in every project. claude mcp add -s user agent-device -- agent-device mcp The agent-device repository includes three agent skills: `agent-device`, `ios-simulator`, and `android-emulator`. The official instruction is `npx skills add callstack/agent-device`, but Claude Code's auto mode safety check blocked it. In that case, clone the repository and copy the three folders under `skills/` into `~/.claude/skills/`. To undo everything, run `claude mcp remove agent-device -s user` and `npm uninstall -g agent-device`, then delete the three skill folders. ## The basic loop The basic loop is: prepare a device, open the app, act, verify, close. Pick the device with `--platform` and `--device`. agent-device boot --platform android --device "<emulator name>" --headless agent-device install /absolute/path/app-debug.apk --platform android --device "<emulator name>" agent-device open <bundle-id> --platform android --device "<emulator name>" --foreground agent-device press @e5 --settle agent-device fill @e11 "text" --settle agent-device wait text "<expected text>" agent-device screenshot /absolute/path/home.png agent-device close `open` reads the screen and returns a ref such as `@e5` for each element, like buttons and text fields. With `--settle`, `press` and `fill` wait until the screen goes quiet and return only what changed, so the agent can pick its next target without rereading the whole screen. Three things commonly go wrong here: * **Refs go stale when the screen changes.** Pick a new ref from the diff returned by each action. * **A screenshot is not verification.** Verify with `wait text` or `wait <selector>`, and keep screenshots as evidence only. * **Give`install` an absolute path.** A relative path resolves against the agent-device daemon's working directory and fails with "App binary not found". For a Flutter app, build the iOS simulator app with `flutter build ios --simulator --debug` and the Android app with `flutter build apk --debug`. ## Flutter gotchas agent-device reads the screen through the accessibility tree. In Flutter, that means only widgets with text show up by default; icon buttons and containers do not. Add `Semantics(identifier:)` or `semanticLabel` to those widgets so a selector can find them. Flutter's `Key` does not appear in the accessibility tree. Labels in the accessibility tree follow the language the device is set to, so the same app reads with different label text depending on that setting. A text-based selector changes with the device language, which makes `identifier` the more stable choice. On iOS, AdMob test banners show up as `platform_view`, `webview`, and `link` nodes. Selectors in replay scripts should avoid those nodes. ## References * agent-device repository: https://github.com/callstack/agent-device * agent-device agent setup docs: https://oss.callstack.com/agent-device/docs/agent-setup * Google Artemis: https://github.com/google/artemis * Maestro Flutter support docs: https://docs.maestro.dev/platform-support/flutter
dev.to
October 24, 2026 at 6:20 PM
アプリ開発の会社Callstackが、本物のスマホでアプリを試すAIを実演した。

利用者がたどる操作を自分で進めるAIだ。それだけなら報告は「テスト成功」の一言で足りる。

このAIは、コードの変更案が出るたびに証拠を置いていく。置き場所は、仲間が変更案を確かめるページだ。中身は画面画像と録画とログ。

置き配でも玄関の写真1枚なのに、録画まで付けるのはやりすぎで好きだ。
October 20, 2026 at 3:13 AM
Now there's good support for Tracy (can inspect a callstack from native C++ up to GDScript) but I don't know if it supports C#

The profilers you named are the best bet for C#

docs.godotengine.org/en/stable/en...
Tracy
Tracy is an Open Source profiler that runs on a wide variety of platforms, including Windows, Linux, and macOS. While it is primarily a tracing profiler, it can also periodically sample data like a...
docs.godotengine.org
October 18, 2026 at 6:12 PM
Your agents work while you sleep. So does your token bill.

Join us live on October 15 📺 clstk.com/4hQnvbK Kewin and Szymon will discuss stalled runs and cost per ticket in React Native software factories. See you there!
Who Left the Agents Running? React Native Software Factories | Callstack
Join Kewin Wereszczyński and Szymon Chmal to explore what can go wrong in a React Native software factory overnight and the controls that keep agents on track.
clstk.com
October 13, 2026 at 5:23 AM
Not sure how I managed to miss this repository!

Offensive techniques using Windows Fibers:

1⃣ PoisonFiber (The first remote enumeration & Fiber injection capability POC tool)
2⃣ PhantomThread (An evolved callstack-masking impl…

— from @ipurple (https://x.com/ipurple/status/2108515099535401057)
GitHub - JanielDary/ImmoralFiber: Two new offensive techniques using Windows Fibers: PoisonFiber...
t.co
October 9, 2026 at 11:17 AM
i already am doing it on the callstack but I don't want to mutate things during reconciliation because then if a component throws I'm left in an incoherent state
October 6, 2026 at 8:43 PM
if you don't need the render to be time-sliced or interruptible you can do the whole render on the callstack and avoid some allocations that way

it can get messy though, there are a shocking number of frameworks that lose react's careful affordances
October 6, 2026 at 8:40 PM
file to get a quick idea of a project (in-house engine) i never touched.
Top most demand i asked, was to draw a diagram of some module, and both gave me fancy looking diagrams, but nothing more informative than if i put a breakpoint and looked into the callstack heriarchy! Yes,..
September 24, 2026 at 12:53 AM
CET-Compliant Callstack Spoofing via Thread Pool Enum Callback Trampolining:

mrtiz.github.io/cet-callstac...

#cybersecurity #programming #infosec #hacking #redteam
September 7, 2026 at 1:17 PM
every day is a wondrous new nightmare
September 3, 2026 at 5:52 AM
This Week In React 295

⚛️
- StyleX
- browser()
- Next.js
- Compiler in Vite/Bun
- Astryx
- Formisch
- React Aria
- TanStack
📱
- C++ API
- Margelo+Callstack
- Observe
- Reanimated
- Teleport
- Nitro
- Screen Transition

🍿 Read thisweekinreact.com/newsletter/295

✍️ @jwr.ski & I
September 2, 2026 at 8:19 PM
CallstackのCEOにリツイされるなど
September 2, 2026 at 1:02 PM
Call Graph Planning: Adapting Effect’s Mental Model for AI
Saat menggunakan AI untuk melakukan koding, biasanya kita melakukan planning terlebih dahulu. Namun, pada planning yang sudah dibuat oleh AI, output planning yang dihasilkan terkadang sulit untuk kita mengerti dan juga terkesan terlalu banyak jargon teknis yang sulit dipahami. Beberapa minggu yang lalu, aku melihat Dillon Mulroy membagikan bagaimana dia menggunakan callstack untuk membuat techical spec Dan aku juga melihat Rin (r17x) menggunakan cara yang sama untuk planning. Melihat penggunaan callstack yang mereka gunakan untuk planning membuatku merasa mudah untuk mereview apa yang sedang di plan oleh AI dan juga kita bisa melihat mock functionality dari callstack tersebut dan error apa yang diproduce. Tdak lama setelah itu, Rin membuat sebuah design thinking yang mengadopsi mental model effect typescript. > _Effect adalah standard library untuk typeScript yang menyediakan runtime khusus agar kode asynchronous, error handling, dan dependency injection menjadi 100% type-safe dan terprediksi._ ## The Design Sin-Think Jika kamu membuka design thinking kamu akan melihat workflow seperti ini: X → Graph → Effect<A, E, R> │ │ │ │ │ │ │ │ │ └─ what each node needs │ │ │ └──── where the graph breaks │ │ └─────── what flows through nodes │ │ │ └─ nodes = functions, edges = data flow │ └─ the problem: what you’re trying to build ### Pemisahan Tiga Channel (A,E,R) Di effect typescript, sebuah function yang memiliki potensi return error selalu diberikan return `Effect<A,E,R>` atau `Effect<A,E>` (jika tidak memiliki dependency). Nah ketiga channel tersebut memiliki sebuah definisi: * _A_(Happy Path): Memaksa AI/human memetakan domain logic dari awal tanpa terkontaminasi error handling dan memberikan sebuah return value * _E_ (Failure Modes): Pengkategorian error menjadi Retry (transient), Escape (recoverable/expected), dan Die (bug/panic) adalah mental model yang universal di bahasa mana pun. * _R_ (Requirements): Sebuah dependency pada function tersebut untuk fleksibilitas testing. ### Menggunakan Call graph untuk Planning Terkadang saat AI melakukan planning, AI sering kali memberikan output teks yang panjang lebar dan disertai jargon sehingga sulit untuk kita pahami. Dengan memberi aturan untuk AI agar menyusun _call graph_ berbasis _A, E, R_ sebelum menulis kode, kita memaksa AI berpikir secara linier dan terstruktur. Alur berpikir _call graph_ pada dasarnya adalah sebuah hirarki kode yang akan ditulis. Production: HTTP Handler → UserService.getUser → UserRepo.findById → PostgresDB Tests: HTTP Handler → UserService.getUser → UserRepoMock Saat AI menyusun _call graph_ , ada beberapa hal menarik yang terjadi: 1. **Pemetaan Alur Data (A)** : Menentukan relasi antar function, data yang dilewatkan, dan struktur output akhir. 2. **Identifikasi _Error Handling_ (E)**: Menentukan titik mana yang dapat menghasilkan _error_ serta penanganannya (_retry_ , _fallback_ , atau _die_) sebelum function dibuat. 3. **Dependency (R)** : Setiap function mendeklarasikan dependency-nya secara terbuka, sehingga tidak ada _hidden dependency_ di dalam kode. Dengan _call graph_ , proses _code review_ terhadap planning AI menjadi lebih cepat. Jika _call graph_ -nya salah atau terlalu rumit, kita bisa langsung memberi feedback di tahap planning sebelum baris kode pertama ditulis. ## Mengadopsi _Call graph Design Thinking_ Mental model _A, E, R_ bersifat agnostik dan dapat diterapkan di luar ekosistem TypeScript. Konsep ini dapat diadaptasi saat menulis kode di bahasa seperti Go, Swift, Kotlin, Dart, dll. Meskipun bahasa pemrograman tersebut tidak memiliki _runtime_ bawaan seperti Effect, prinsip dasarnya tetap dapat diimplementasikan: * **Go** : **A** direpresentasikan sebagai _return value_ , **E** sebagai _custom error_ untuk memisahkan _retryable error_ dan _domain error_ , dan **R** dipassing secara eksplisit melalui _struct_ atau _interface injection_. * **Kotlin** : **A** dan **E** bisa menggunakan `Either<E, A>` jika pakai library Arrow-kt atau sealed class/interface sebagai typed error, sedangkan **R** dipassing menggunakan constructor injection atau function parameter. * **Swift** : **A** dan **E** bisa menggunakan `Result<T, E>` atau async throws ber-tipe, sedangkan **R** menggunakan protocol-based dependency injection. * **Dart / Flutter** : **A** dan **E** bisa menggunakan `Either<E, A>` jika pakai library fpdart, dan **R** dipassing menggunakan constructor injection atau function parameter. Untuk mengadopsi _design thinking_ ini ke dalam _workflow_ AI (seperti Cursor, Claude, atau LLM lainnya), kamu bisa menggunakan contoh prompt ini In this session we are going to build a "[YOUR IDEA]" and this app has features: [Breakdown The MVP] The tech stack is: - Flutter Use best practice coding in Flutter and adopt this design thinking: "[ATTACH GIST DESIGN THINKING URL/CONTENT]" Actually this design thinking is based on the Effect TypeScript mental model, but we can adapt its core concepts (A, E, R channels, call graph planning, typed errors, clean dependency separation) into Flutter functional programming. Please port the design thinking into effect-flutter skills before planning the architecture. Prompt diatas hanyalah sebuah contoh saja dan aku sudah coba menerapkan ini pada projek swift. Ini adalah output yang dihasilkan claude saat melakukan planning ## Penutup Dengan mengadopsi cara ini, AI akan memetakan bagian happy path, error yang bisa direcovery atau tidak, kebutuhan dependency dari sebuah function/class, dan kita bisa memverifikasinya dengan mudah karena rancangan sistemnya lebih jelas.
dev.to
August 31, 2026 at 9:11 AM