Android Feed

← The 5 stories that mattered

Everything collected ·

All 396 items collected for this issue

New here? How to read this issue
  1. Read the picks. Each tab opens with 4–5 stories ranked by how much you’d regret missing them. That takes about 3 minutes. “Start the read” shows one story at a time, and “Listen” reads them aloud.
  2. Stop at “You’re caught up”, or go deeper. Everything after it is the full list of what was collected today, folded away.
  3. Android comes first. Product & AI covers what builders are launching and discussing.
Stable · Beta · Alpha
Release stages. Stable is safe to ship, beta’s API is frozen but may still have bugs, alpha is experimental.
AndroidX
Google’s Jetpack libraries (Compose, Paging, Room…), released separately from Android itself.
AOSP insider
Code changes merged into Android’s open source repository. They often show an API before it’s announced.
Lanes
Where an Android item comes from: official, AOSP insider, experts, open-source library releases, leaks and ecosystem, community.
Adopt now · Watch · Rumour
What to do about it, and how sure the source is.
Dotted underline
An inferred claim: Claude’s reading, not something the source says.

Collected automatically each morning from 37 sources. Notes are written by Claude. Links show what was trending, not endorsements.

Android5Product & AI4
Start the Android read5 stories, ranked by what you’d regret missing · 3 min read

  1. Android Studio Rabbit 2 is stable and runs any coding agent

    You no longer have to choose between your preferred agent and the IDE's own tools, because the agent you already pay for now works inside Android Studio.

    Google released Android Studio Rabbit 2 to the stable channel at Android Dev Summit. Its main feature is Bring Your Own Agent. You can run Claude Agent, Codex, Junie, Google Antigravity or any of over 40 agents that speak the open Agent Client Protocol inside the IDE, with access to the emulator, builds and Logcat. It also adds a skills manager, agent-assisted LeakCanary fixes, AI change review and stable XR emulators.

    Android Developers Blog · OfficialRead the sourceProduct notes

  2. Android CLI gets Compose previews, device streaming and agent plugins

    Agents working outside the IDE can finally see and test what they built, which cuts both cost and the number of broken screens they hand back.

    The second Dev Summit post covers tools. The Android command line tool can now render Jetpack Compose previews without an emulator, stream a real device from Google's lab, analyse heap dumps on your computer, and score an app bundle's R8 shrinking the way Play does. Official plugins for Claude Code and Codex install it in one command. Google says agents using terminal previews finished UI tasks 77 percent faster with far fewer tokens.

    Android Developers Blog · OfficialRead the sourceProduct notes

  3. XREAL Aura is the first Android XR glasses, and the XR SDK is stable

    Android XR now has a consumer device and stable APIs on the same day, so adding a spatial layout to an existing app is a real option rather than a bet.

    XREAL announced Aura, the first wired glasses that run Android XR, Google's version of Android for headsets and glasses. It ships in November, runs existing Android apps, and launches with over 150 experiences made for XR. The same day, the core Jetpack XR libraries, SceneCore, ARCore for Jetpack XR and XR Runtime, reached version 1.0 stable, and Android Studio's XR emulators left preview.

    HN front page · CommunityRead the sourceDiscussionProduct notes

  4. Haze 2.0 brings glass effects to Compose Multiplatform

    If your designers want the glass look that iOS and Android are converging on, this is the one library that gives it to you across platforms with the native fast path when available.

    Chris Banes released Haze 2.0, the blur library for Jetpack Compose and Compose Multiplatform. It adds Glass, a frosted-glass look, a new effects API for combining effects, performance improvements, and experimental support for Android's native backdrop, the system blur added in recent Android versions. A companion post explains when an agent task needs a plan, using two Haze tasks as examples.

    Chris Banes · ExpertRead the sourceProduct notes

  5. LiteRT in Play services gets faster GPU inference and smaller apps

    On-device models get faster and your APK gets smaller without shipping a new runtime yourself, which lowers the cost of adding local AI to an app.

    Google's developer blog describes an update to LiteRT, the on-device machine learning runtime, as delivered through Google Play services. It now runs models on the GPU through a new engine called ML Drift, about 1.4 times faster than before, and adds a CompiledModel API. Because the runtime ships with Play services, apps that switch stop bundling it and get smaller.

    Google Developers Blog · OfficialRead the source

Also worth knowing

Releases, policy and leaks an Android engineer should know about this week.

  • Agentic and Flexible: A next era of Android development

    Android Dev Summit made several agent and adaptive pieces official: AppFunctions is in beta as Android's tool-calling layer exposed through Android MCP, A2UI for Compose and an ML Kit Agentic API with Gemini Nano 4 are in alpha, AppState is in alpha, Navigation 3 1.2 is stable with deep links and richer predictive back, and the core Jetpack XR libraries are 1.0 stable. Google also said Play will look more closely at foldable support next year.

    Any app that wants to be callable from Gemini, and every app that ships on foldables, Googlebook or glasses.

    Adopt nowOfficial

  • A more reliable way to add and edit contacts on Android

    Apps targeting the upcoming Android release lose direct contact writes through WRITE_CONTACTS. Use ACTION_INSERT, ACTION_EDIT or ACTION_INSERT_OR_EDIT, which need no permission, or apply for the new MANAGE_CONTACTS special access arriving in Android 17 QPR2 if managing contacts is the app's core job.

    Contact sync, backup, caller ID and CRM apps; custom MIME types in insert extras are ignored by the system editor.

    Adopt nowOfficial

  • kotlinx.serialization 1.12.0

    kotlinx.serialization 1.12.0 is stable, built on Kotlin 2.4.10, and fixes unnecessary leading whitespace in output.

    Every Kotlin app using kotlinx.serialization; a safe update.

    Adopt nowOfficial

  • 1.13.0-beta01

    Compose Multiplatform 1.13.0-beta01 is out; desktop libraries using the experimental DragAndDropTransferAction must be recompiled because of a binary-incompatible change.

    Library authors shipping Compose Multiplatform desktop code with drag and drop.

    WatchOfficial

  • People APIs hidden until API review

    Gerrit hides a new androidx.people library until API review. It lines up with the People Directory that Google says arrives in Android 17 QPR2 for apps that add Call with or Message with actions instead of writing contacts.

    Messaging and calling apps that today insert their own entries into the user's contacts.

    WatchInsider

Go deeper: every Android item, by lane

Official & releases

58Android Dev Summit day: Android Studio Rabbit 2 is stable with any ACP agent inside it, the Android CLI renders Compose previews and streams real devices, AppFunctions is beta, Navigation 3 1.2 and the Jetpack XR libraries are stable, Compose 1.13 beta02 is the last stop before stable, and LiteRT in Play services gets faster GPU inference.

AndroidX library releases

Stable · safe to ship · 8

Beta · API frozen, try it before it ships · 13

Alpha · experimental, APIs may change · 22
Show 7 more

AOSP / AndroidX insider

24Gerrit shows a hidden androidx.people library waiting for API review, AppState getting DataStore storage, Compose Remote learning intrinsic sizes and unknown-operation policies, and an agent config for Compose performance reviews checked into AndroidX itself.
  • Android XR: helper waits for the spatial scenenew

    Add awaitSpatialScene and embeddedWindowLeashes to SpatialSceneHelper

    SpatialSceneHelper gains awaitSpatialScene and embedded window leashes, groundwork for placing ordinary app windows inside XR scenes now that SceneCore is 1.0.

  • Compose: remember a style resolvernew

    Add `rememberStyleResolver()`

    A remembered style resolver points to the Formatted Text work in Compose 1.13, where styles are resolved once and reused across recompositions.

  • People APIs hidden until API reviewnew

    Hide androidx.people APIs until API review

    A new androidx.people library exists and is waiting for the API council; it matches the People Directory promised for Android 17 QPR2.

  • Glance Wear: fallback set trimmed to 1.6.0 operationsnew

    Remove post-1.6.0 ops from Glance Wear SAFE_FALLBACK_SUPPORTED_OPERATIONS

    Wear widgets keep a safe list of operations older players understand; newer ones are removed so widgets degrade cleanly on old watches.

  • Lazy lists: slide animations for itemsnew

    Add Slide `EnterExitTransition` to `animateItem`

    animateItem gains slide enter and exit transitions, extending item animation beyond fade and placement.

  • Compose: an agent config for performance reviewsnew

    Add Compose performance review agent config

    AndroidX itself checks in an agent configuration for reviewing Compose performance, mirroring the jetpack-compose-performance skill announced at Dev Summit.

  • Embedded player: policy for unknown operationsnew

    Add unknown-operation policy to embedded RcPlayer

    Compose Remote's embedded player decides what to do with operations it does not recognise, needed so documents from newer apps still render on older hosts.

  • Embedded player: optional featuresnew

    Add optional features to embedded RcPlayer

    The embedded player can declare optional capabilities, so a remote document can adapt to what the host supports.

Show 16 more

Experts

17Read Chris Banes on Haze 2.0 and on when an agent task needs a plan, skydoves' three-part series on how Compose compiles, remembers and recomposes, and Kt. Academy on Flow bugs that look correct.
Show 9 more

Library releases

7kotlinx.serialization 1.12.0 is a safe stable update, Metro 1.5.0 adds linuxX64 and a diagnostic-level option, Haze 2.0 is a major with a new effects API, OkHttp 5.5.1 and 4.12.1 are patch releases worth taking after this week's CVE write-up, Compose Unstyled 2.10.2 fixes focus inside hosts, Roborazzi 1.77.0 fixes configuration cache reuse, and Decompose 3.6.0 stays in beta.
  • Compose Unstyled 2.10.2

    Compose Unstyled 2.10.2 fixes dialogs, sheets and drawers inside a Host and keeps Tab focus in the topmost layer.

  • Metro 1.5.0

    Metro 1.5.0 adds linuxX64 for the ViewModel Compose module and a diagnostic-level compiler option.

  • OkHttp 5.5.1

    OkHttp 5.5.1, a patch release.

  • OkHttp 4.12.1

    OkHttp 4.12.1, a patch release for the 4.x line.

  • kotlinx.serialization 1.12.0

    kotlinx.serialization 1.12.0 is stable on Kotlin 2.4.10.

  • Decompose 3.6.0-beta02

    Decompose 3.6.0-beta02 adds Linux targets and fixes RTL predictive back directions.

  • Roborazzi 1.77.0

    Roborazzi 1.77.0 fixes configuration cache reuse for verify and compare tasks.

Leaks & ecosystem

23XREAL Aura ships in November as the first Android XR glasses, Gemini and Search pass 50 connected apps, Google says Play will reward well-adapted foldable apps, and teardowns show a Gemini memory page, Wallet payments while locked and Find Hub alert fixes.
Show 15 more

Community

40Hacker News is busy with XREAL Aura and iOS firmware on Android, GitHub's Kotlin trending is led by Vela offline navigation, GKD and v2rayNG, and Android Weekly covers AppFunctions, Compose smoothness, Gemini Nano's token budget and shipping a Gemma model through Play.
Show 32 more
Start the Product & AI read4 stories, ranked by what you’d regret missing · 2 min read

  1. Opera lets websites use the AI on your computer

    If browsers become the AI runtime, any site can add AI features at no per-request cost, which changes who can afford to build them.

    WebAI by Opera topped Product Hunt. It lets a website call AI models that run on the visitor's own computer, through the browser, instead of sending requests to a server. For the site owner that means no API bill and no user data leaving the machine. It only works in Opera for now, so reach depends on whether other browsers adopt a similar standard.

    Product Hunt · New ideaRead the sourceProduct notes

  2. A benchmark puts agents on 50 real tax assignments

    Benchmarks with checkable answers in real professions will decide which claims about agents replacing work survive, so watch who publishes them and how they score.

    Dayjob: Tax evaluates AI agents on 50 professional tax assignments of the kind a junior accountant would do, and grades whether the work is right. It drew about 220 points and 140 comments on Hacker News, with debate on how close agents are to office work. On the same day Google introduced Android Bench 2.0, which grades agents on Android tasks that take an engineer days.

    HN front page · Worth a debateRead the sourceDiscussionProduct notes

  3. DeepTutor, a tutor that remembers you, tops GitHub growth

    Memory across sessions is the feature general chat assistants structurally lack, and education is the first place people are willing to pay for it.

    DeepTutor, an open-source project from a Hong Kong university lab, calls itself lifelong personalized tutoring. It keeps a memory of what each learner studied and adapts explanations over time instead of starting fresh each chat. It added about 2,400 stars today, more than any other repository, and sits near 48 thousand overall.

    GitHub Trending · Fastest risingRead the sourceProduct notes

  4. Confetti builds hardware projects from plain-English prompts

    Hardware still needs a specialist at every step, so a plain-language front end has more room there than in software, if it can prove the boards work.

    Confetti, the most upvoted Y Combinator launch of the day, lets you start a hardware project from a conversation. You describe the device you want and it works through the design with you, in the way coding agents work through software. The launch does not yet say what gets manufactured or how designs are verified, and on Hacker News a factory-scale robot that builds circuit boards on demand drew attention the same day.

    YC Launches · New ideaRead the sourceProduct notes

Go deeper: themes, product notes, every source

Themes across sources

6Agent-ready Android · Coding-agent control surfaces · On-device AI · Anti-slop design skills · Verifying agent work · Android XR glasses

Ideas showing up in more than one place today. The line counts items per day over two weeks.

Agent-ready Android

Rising · 8 items from 5 sources

At Android Dev Summit, Google made the pieces assistants need official: AppFunctions, the app-level tool-calling layer, is in beta, A2UI for Compose lets agents answer with an app's own components in alpha, and a Connected Apps platform plus an ML Kit Agentic API with Gemini Nano 4 round it out. Android Authority reports Gemini and Search now have over 50 connected apps, and Android Weekly carries AppFunctions and on-device agent explainers.

Why now. Gemini is becoming the system agent on the phone, so apps need typed entry points and UI that an agent can call and render.

Agentic and Flexible: A next era of Android developmentAndroid Developers Blog

Coding-agent control surfaces

Rising · 12 items from 4 sources

Android Studio Rabbit 2 goes stable with Bring Your Own Agent, running Claude, Codex, Junie, Antigravity or any of over 40 agents that speak the Agent Client Protocol, and the Android CLI gains official Claude Code and Codex plugins. On the product side, JetBrains' Thinkrail keeps climbing on GitHub, and Product Hunt lists Zest, Yardlet, Mudroom, Trellis and Paddock, each a different way to launch, watch or isolate agents. On Hacker News, a post on the Agent SDK and a complaint about Claude Code's status line tokens both drew long threads.

Why now. Open protocols such as ACP and MCP let one IDE or terminal host many agents, so the competition has moved to who controls the surface they run in.

Your favorite agents are now integrated in Android StudioAndroid Developers Blog

More on this

On-device AI

Rising · 11 items from 5 sources

Opera's WebAI, the top Product Hunt launch, lets websites use AI running on the visitor's own computer, and Hob pitches an open-source, on-device alternative to Claude and ChatGPT. Liquid AI's LFM3 family is on Product Hunt and Hugging Face, NVIDIA's Nemotron 3.5 Bolt fits in 4 GB, and on Android, LiteRT in Google Play services gets about 1.4x faster GPU inference while Android Weekly covers Gemini Nano's token budget and shipping a 584 MB Gemma model through Play.

Why now. Small open models now handle everyday tasks on a phone or laptop, and the runtimes to serve them on GPUs and NPUs are shipping inside the platform.

WebAI by OperaProduct Hunt

Anti-slop design skills

Rising · 7 items from 2 sources

Design taste packaged as agent skills is again the fastest-growing corner of GitHub: ai-design-skills added about 2,100 stars today, Hallmark about 2,100 and Impeccable about 1,500, with Superpowers past 300 thousand. The format is spreading beyond design, with Expo publishing skills for its tooling, a Product Hunt launch that has a coding agent design App Store screenshots, and Google shipping over 30 Android skills.

Why now. Skills became a shared format across Claude Code, Cursor, Codex and now Android Studio, so one file of rules reaches every agent a team uses.

elayadesign/ai-design-skillsGitHub Trending

More on this

Verifying agent work

Rising · 7 items from 4 sources

Dayjob: Tax, a benchmark that gives agents 50 real professional tax assignments, drew a 140-comment Hacker News debate, and Google's Android Bench 2.0 grades agents on multi-day Android tasks. Products follow: Revix AI reviews code against a repo's own rules, BotGauge calls itself a trust layer for agents, Dedica checks whether agents can use your website, GrayPass stops agents after login, and Rystic gives them local copies of Slack and Stripe to practise on. A Dev.to post lists four ways a green test can lie.

Why now. Agents now take real actions and write large changes, so the bottleneck has moved from producing work to measuring and trusting it.

Dayjob: Tax – evaluating AI agents on 50 professional tax assignmentsHN front page

Android XR glasses

New · 9 items from 6 sources

XREAL Aura, the first wired glasses running Android XR, launched with a Dev Summit announcement, 150 made-for-XR experiences and a November ship date, and reached both Hacker News front page and the Android feed. The same day, the core Jetpack XR libraries, SceneCore, ARCore for Jetpack XR and XR Runtime, reached 1.0.0 stable, XR emulators went stable in Android Studio, and Gerrit shows work on embedding app windows in spatial scenes.

Why now. A consumer device is finally shipping, so the XR APIs that have been in alpha for a year now carry stability promises.

The first wired XR glasses powered by Android XRHN front page

Fading: World models (last seen 2026-10-28) · Agent supply-chain safety (last seen 2026-10-27) · Agents as security auditors (last seen 2026-10-26) · Open voice models (last seen 2026-10-26) · Agent sandboxes (last seen 2026-10-25)

Product notes

17Who each product is for, the problem, why now, its wedge and how it makes money. Today’s picks first.

Plain text: the source says soDotted: inferred

WebAI by Opera

productProduct Hunt#1 on Product HuntHype flag

What
WebAI by Opera lets websites use AI models running on the visitor's own computer, the top Product Hunt launch.
For whom
Web developers who want AI features without paying per request or sending user data to a server.
Problem
Every AI feature on a site today means an API bill and a privacy disclosure; sites cope by limiting usage or skipping the feature.
Why now
Small models now run well on laptops, and browsers can expose them to pages the way they expose cameras.
Wedge
The browser as the AI runtime, so the site needs no model hosting.
Money
Opera gains a reason to pick its browser; no direct charge stated.
Like
Chrome built-in AI APIs · WebLLM · Edge Copilot

Depends on sites adopting an Opera-specific capability; reach is bounded by Opera's share unless it becomes a web standard.

LessonMoving the cost of a feature onto hardware the user already owns changes who can afford to offer it.

Dayjob: Tax – evaluating AI agents on 50 professional tax assignments

discussionHN front page217 pts · 140 comments

What
Dayjob: Tax, an evaluation of AI agents on 50 professional tax assignments, with about 220 points and 140 comments on Hacker News.
For whom
Firms and labs deciding whether agents can take on professional work, and the professionals who would review it.
Problem
Coding benchmarks say little about office work, so claims about agents replacing professionals have no shared yardstick.
Why now
Agents can now run long multi-step jobs with documents and tools, so realistic assignments are testable.
Wedge
One profession with clear right answers, tax, rather than a broad office benchmark.
Money
Not stated; benchmarks of this kind often lead to consulting or evaluation services.
Like
SWE-bench · GDPval · Android Bench 2.0 · Terminal-Bench

LessonA benchmark in a field with verifiable answers settles arguments that marketing cannot.

HKUDS/DeepTutor

repoGitHub Trending★ 2,371 today

What
DeepTutor from HKUDS, pitched as lifelong personalized tutoring, about 48 thousand stars after adding about 2,400 today.
For whom
Learners who want an assistant that remembers what they studied and adapts over time.
Problem
Chat assistants forget each session, so learners repeat context and get generic explanations.
Why now
Agent memory and retrieval are cheap enough to keep a per-learner model over months.
Wedge
Lifelong memory as the distinguishing claim rather than better answers.
Money
Open source, with a hosted site; no pricing stated.
Like
Khanmigo · ChatGPT study mode · Gemini Guided Learning

LessonPick the one property the incumbents structurally lack, here memory across sessions, and name the product after it.

Confetti: Build hardware projects using plain-English prompts

productYC Launches90 votesHype flag

What
Confetti, a YC launch for building hardware projects from plain-English prompts, the most upvoted launch of the day.
For whom
Makers and engineers who can describe a circuit but struggle with schematics, part selection and board layout.
Problem
Hardware design needs several specialist tools and vocabularies; beginners cope with dev boards and forum threads.
Why now
Agents can now plan across tools, and on-demand PCB factories such as The Grid make small runs cheap.
Wedge
A conversation as the starting point, lowering the first step for non-specialists.
Money
Likely subscription plus a cut of board and part orders.
Like
Flux.ai · JLCPCB · Arduino Cloud

Prompt-to-hardware is a large claim; the launch does not say what gets manufactured or verified.

LessonA vertical that still needs human experts at every step is where a plain-language front end has the most room.

Your favorite agents are now integrated in Android Studio

articleAndroid Developers BlogAndroid

What
Android Studio Rabbit 2, now stable, with Bring Your Own Agent: pick Claude Agent, Codex, Junie, Google Antigravity or any of over 40 agents that speak the Agent Client Protocol, plus stable XR emulators, a skills manager, LeakCanary fixes with an agent and AI change review.
For whom
Android teams that already pay for a coding agent and want it inside the IDE with access to the emulator, build system and Logcat.
Problem
Agents in a terminal cannot see previews, run the emulator or read Logcat, so developers shuttle context by hand.
Why now
ACP gives IDEs one protocol for many agents, so Google can host rivals instead of forcing Gemini.
Wedge
Letting developers keep the agent they already use rather than switching to Google's.
Money
Free IDE; Google earns through Antigravity, Gemini models and Gemini Enterprise for companies.
Like
Zed with ACP agents · JetBrains Thinkrail · Cursor · VS Code Copilot agent mode

LessonWhen users already chose a tool, hosting it beats competing with it, and the protocol becomes the product.

What's new in Android developer tools and performance at Android Dev Summit

articleAndroid Developers BlogAndroid

What
The Android CLI gains Compose previews from the terminal, remote real-device streaming, a Device Hub, LeakCanary heap analysis on the host, an analyze command for R8 scores and official Claude Code and Codex plugins. Google says agents using terminal previews finished UI tasks 77 percent faster with up to three times fewer tokens in internal tests.
For whom
Developers running coding agents outside Android Studio who need the agent to see what it built.
Problem
An agent with no preview or device must build, deploy and screenshot an APK to check UI, which is slow and costly.
Why now
Agents now do whole UI tasks, so the tool loop around them determines cost and speed.
Wedge
Headless Compose previews, something only Google's tooling can render faithfully.
Money
Free tooling; device streaming runs through Firebase, which is paid at scale.
Like
Maestro · Xcode command line tools · Expo skills

LessonMeasure the agent loop in tokens and minutes; a tool that shortens it sells itself with one number.

Agentic and Flexible: A next era of Android development

articleAndroid Developers BlogAndroid

What
Google's Dev Summit overview: AppFunctions in beta, A2UI for Compose in alpha, a Connected Apps platform for publishing to Gemini, an ML Kit Agentic API with Gemini Nano 4, AppState in alpha, Navigation 3 1.2 stable, stable Jetpack XR libraries, and a note that Play will look harder at foldable support next year.
For whom
Every Android app team deciding what to expose to Gemini and how to stay adaptive across phones, foldables, laptops and glasses.
Problem
Assistants cannot act inside apps without typed functions, and apps break across form factors without a shared state model.
Why now
Gemini is the system agent and Googlebook and XR glasses are shipping, so the platform needs both agent hooks and adaptive UI at once.
Wedge
AppFunctions as the single tool-calling layer, exposed through Android MCP.
Money
Platform play: more actions in Gemini and more apps across Google devices.
Like
Apple App Intents · MCP servers · Alexa skills

LessonA platform earns adoption by giving developers one integration that reaches every surface it owns.

JetBrains/thinkrail

repoGitHub Trending★ 1,068 today

What
Thinkrail, JetBrains' lightweight real IDE for vibe coding with the pi agent, which customises itself around how you work; about 5,300 stars after adding about 1,100 today.
For whom
Developers who want an agent-first editor but miss real IDE features such as indexing and refactoring.
Problem
Agent-first editors are thin wrappers over a chat box, while full IDEs bolt agents on the side.
Why now
Agents write most of the code, so the IDE's job shifts to steering and review.
Wedge
An IDE that reshapes itself to the user instead of shipping a fixed layout.
Money
Likely a funnel into JetBrains' paid IDEs and AI subscriptions.
Like
Cursor · Zed · Android Studio Rabbit 2 · Windsurf

LessonAn incumbent can enter a new category with a separate small product rather than retrofitting the flagship.

Hob

productProduct Hunt#4 on Product HuntHype flag

What
Hob, an open-source, on-device alternative to Claude and ChatGPT.
For whom
People who want a chat assistant that keeps everything on their machine.
Problem
Hosted assistants send every message to a company's servers; local alternatives are fiddly to set up.
Why now
Open models such as LFM3, Gemma and Qwen are good enough for daily chat on consumer hardware.
Wedge
Positioning as a direct replacement for the two best-known assistants.
Money
Open source, no model yet.
Like
Jan · LM Studio · Ollama · Off Grid AI

Named against frontier assistants while running small local models; quality parity is unproven.

LessonComparing yourself to the category leader buys attention, but you must then meet the expectation it sets.

elayadesign/ai-design-skills

repoGitHub Trending★ 2,139 today

What
ai-design-skills, a collection of design skills for coding agents, about 11,300 stars after adding about 2,100 today, the most of any repo.
For whom
Developers using agents to build interfaces who want them to follow design rules rather than defaults.
Problem
Agents produce generic layouts and the same tells every time; teams cope by hand-polishing or writing long prompts.
Why now
Skills became a standard extension format across agents, so a design playbook can be installed rather than pasted.
Wedge
A curated collection rather than one opinionated skill, so teams pick what fits.
Money
Open source, no model yet.
Like
Hallmark · Impeccable · Superpowers

LessonWhen a format spreads, the first good collection in a popular niche grows faster than any single tool.

Hopper – Fast, open-source inference for real-time agents

productYC Launches46 votes

What
Hopper, a YC launch that trains open-weight models on your production traces, then serves and improves them against live traffic for real-time agents.
For whom
Teams running latency-sensitive agents who pay frontier-model prices for narrow, repeated tasks.
Problem
Frontier models are slow and costly for a fixed task; fine-tuning and serving an open model yourself is a project.
Why now
Open-weight models are close enough that traces from a frontier model can teach a smaller one the job.
Wedge
Using traffic you already produce as the training set, so there is no data project.
Money
Inference pricing undercutting hosted frontier models.
Like
OpenPipe · Fireworks fine-tuning · Together AI

LessonTurning the customer's own exhaust into the product input removes the setup step that kills most ML tooling.

Show HN: SlideBySide – Present slides without sharing your screen

productShow HN65 pts · 26 comments

What
SlideBySide, the top Show HN, presents slides to everyone on their own device without screen sharing, with co-hosts able to control the current slide.
For whom
Remote presenters and meetups where screen sharing is laggy, low resolution or blocked.
Problem
Screen sharing sends video of slides; viewers cannot zoom, copy or follow along on a phone.
Why now
Browsers and real-time sync make a shared pointer into a deck trivial compared with streaming pixels.
Wedge
Replacing one habit, share screen, with a link.
Money
Not stated.
Like
Pitch · Google Slides present mode · Slidev

LessonSending state instead of pixels is a recurring win; look for places where people still stream a screen.

LFM3 by Liquid AI

productProduct Hunt#13 on Product Hunt

What
LFM3 by Liquid AI on Product Hunt: six 2-billion-class models for phones, laptops and robots; the 1.7B model also trends on Hugging Face.
For whom
Developers who need a model that fits on a device for one focused task.
Problem
Large hosted models are costly and need a network; small models often handle only text.
Why now
Devices ship NPUs and runtimes such as LiteRT, so small specialised models have somewhere to run.
Wedge
A family of same-sized models for different jobs rather than one general model.
Money
Open weights plus licensing and custom models for companies.
Like
Gemma 4 · Qwen small models · Gemini Nano · Nemotron Bolt

LessonA model family sized for a device class gives developers a budget to design around.

Dedica

productProduct Hunt#12 on Product Hunt

What
Dedica checks whether AI agents can use your website.
For whom
Site owners who expect agents, not people, to book, buy and browse on their behalf.
Problem
Sites built for humans break agents with modals, hidden buttons and bot blocks, and owners have no way to see it.
Why now
Browser-using agents are common enough that agent readiness is becoming a new kind of SEO.
Wedge
A diagnostic score first, fixes later.
Money
Likely freemium audits with paid monitoring.
Like
Lighthouse · Notra for AI search · Pomerado

LessonEvery new kind of visitor creates a market for tools that test the site through that visitor's eyes.

The first wired XR glasses powered by Android XR

discussionHN front page, HN · AndroidAndroid54 pts · 48 comments

What
XREAL Aura, the first wired XR glasses powered by Android XR, discussed on Hacker News and the Android feed, shipping in November with over 150 made-for-XR experiences.
For whom
Early adopters wanting Android apps on glasses, and developers deciding whether Android XR is worth a target.
Problem
Standalone headsets are heavy and expensive; wired glasses offload compute to a puck or phone.
Why now
The Jetpack XR libraries reached 1.0 stable the same day, so apps built now will not break.
Wedge
Running existing Android apps on day one rather than needing an XR-only catalogue.
Money
Hardware sales with a 200 dollar launch discount reported by 9to5Google; Google earns through Play and Gemini.
Like
Meta Ray-Ban Display · Apple Vision Pro · Samsung Galaxy XR

LessonA new form factor sells faster when the existing app catalogue works on it unchanged.

Haze 2.0

articleChris BanesAndroid

What
Haze 2.0, Chris Banes' blur library for Compose Multiplatform, adds Glass, a new effects API, performance work and experimental Android native backdrop support.
For whom
Compose and Compose Multiplatform teams that want frosted-glass and blur effects without writing shaders.
Problem
Real-time blur on Android has been slow or device-specific; teams cope with static images or skip the effect.
Why now
Android 16 and Compose 1.13 add native backdrop and progressive blur, so a library can wrap them with fallbacks.
Wedge
One API across Android, iOS and desktop with the platform fast path when available.
Money
Open source, no model yet.
Like
Compose progressive blur · Cloudy · iOS UIVisualEffectView

LessonA library that wraps a new platform feature with fallbacks lets teams adopt it years before their minimum SDK does.

ruvnet/RuView

repoGitHub Trending★ 557 todayHype flag

What
RuView turns ordinary WiFi signals into spatial sensing, presence detection and vital-sign monitoring without cameras, about 101 thousand stars.
For whom
Home automation and elder-care builders who want occupancy and breathing data without filming people.
Problem
Cameras raise privacy objections and radar hardware is extra cost; WiFi is already in every room.
Why now
Channel state information from commodity routers and small models to decode it have both become accessible.
Wedge
Privacy, no pixels, as the headline rather than accuracy.
Money
Open source, no model yet.
Like
Origin Wireless · Aqara presence sensors · mmWave radar modules

Vital-sign monitoring from WiFi is a research-grade claim; accuracy on real routers is not shown in the listing.

LessonReusing infrastructure that is already installed is the cheapest hardware wedge, but medical-sounding claims need evidence.

By source

11Everything collected today from GitHub Trending, Product Hunt, Show HN, HN front page, YC Launches, HF trending models, HF trending spaces, HF daily papers, Lobsters, Dev.to top, Techmeme, in each source’s own order.

GitHub Trending

14
  • JetBrains/thinkrailcard

    Vibe code with pi in a lightweight, real IDE that customises itself around the way you work — The Vibe You Need

    ★ 1,068 today

  • elayadesign/ai-design-skillscard

    ★ 2,139 today

  • obra/superpowers

    Superpowers, an agentic skills framework and development method, now past 307 thousand stars.

    ★ 1,095 today

  • HKUDS/DeepTutorcard

    DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/ .

    ★ 2,371 today

  • macro-inc/macronew

    Macro, a unified team workspace where email, chat, docs, tasks and agents share one AI memory.

    ★ 117 today

  • Nutlope/hallmark

    Hallmark, the anti-slop design skill, adds about 2,100 stars and passes 40 thousand.

    ★ 2,110 today

  • pbakaus/impeccable

    Impeccable, the design vocabulary for coding agents, adds about 1,500 stars and sits near 93 thousand.

    ★ 1,472 today

  • makecindy/cindy

    Cindy, an open-source agent pitched as working out of the box, keeps trending.

    ★ 728 today

Show 6 more
  • nightly-labs/openbot

    openbot, a local-first desktop workspace for persistent agent teammates across Codex, Claude and Grok.

    ★ 1,054 today

  • togg53192-cmd/jailbreaks

    A collection of LLM jailbreak prompts, trending but not something to build on.

    ★ 1,477 today

  • hmjz100/LinkSwiftnew

    A browser script for direct download links from eight Chinese cloud drives.

    ★ 1,616 today

  • expo/skillsnew

    Expo publishes official agent skills for Expo projects and EAS, the mobile-framework version of Android skills.

    ★ 90 today

  • ruvnet/RuViewnewcard

    π RuView turns commodity WiFi signals into real-time spatial intelligence, vital sign monitoring, and presence detection — all without a single pixel of video.

    ★ 557 today

  • xiaolai/the-craft-of-selfteachingnew

    A Chinese book on the craft of teaching yourself, trending again.

    ★ 148 today

Product Hunt

49
  • WebAI by Operanewcard

    Let websites use AI running on your computer

    #1 on Product Hunt

  • Dotversenew

    Dotverse gives the Dot companion app a crew of characters.

    #2 on Product Hunt

  • AirPosturenew

    AirPosture adds a Mac app and a voice coach in your AirPods for posture.

    #3 on Product Hunt

  • Hobnewcard

    Open-source, on device alternative to Claude and ChatGPT

    #4 on Product Hunt

  • throxynew

    Cold calls made in the prospect's own language, sold as delivered pipeline.

    #5 on Product Hunt

  • OpenLLMnew

    OpenLLM promises to take your AI tokens to any provider.

    #7 on Product Hunt

  • TONKOnew

    A pocket money app that teaches kids about money.

    #8 on Product Hunt

  • Genie 4new

    Genie 4 on Product Hunt, DeepMind's steerable generated worlds from a prompt.

    #9 on Product Hunt

Show 41 more
  • bytemailnew

    A native Mac email client that lives in the notch and strips trackers.

    #10 on Product Hunt

  • BotGaugenew

    BotGauge calls itself the trust layer for agents, another entry in verifying what agents do.

    #11 on Product Hunt

  • Dedicanewcard

    Check if AI agents can use your website

    #12 on Product Hunt

  • LFM3 by Liquid AInewcard

    Six 2B-class models for phones, laptops and robots

    #13 on Product Hunt

  • Fraud Blockernew

    AI ad fraud protection for Google and Meta campaigns.

    #14 on Product Hunt

  • Tulunew

    Autonomous marketing operations for small businesses.

    #15 on Product Hunt

  • App Store Screenshotsnew

    Your coding agent designs the app store screenshots, a skill aimed at the last mile of shipping an app.

    #16 on Product Hunt

  • NotiKeepnew

    Notification history for iPhone kept on the device, a feature Android has had for years.

    #17 on Product Hunt

  • Basedash View-only membersnew

    Basedash adds view-only members for sharing dashboards.

    #18 on Product Hunt

  • Zestnew

    Zest: hold a key, flick, and an agent does the dev chore, a gesture as the agent launcher.

    #19 on Product Hunt

  • RageLayernew

    Lets visitors smash your website and sweeps it back, a marketing toy.

    #20 on Product Hunt

  • OneDeploynew

    OneDeploy runs open-source apps on your own server with no per-seat fees.

    #21 on Product Hunt

  • Congress Tradesnew

    Shows how members of Congress trade and whether copying them pays.

    #22 on Product Hunt

  • Yardletnew

    Yardlet, a Mac app that organises the coding agents you already use.

    #23 on Product Hunt

  • Open For salesnew

    A sales campaign from one prompt.

    #24 on Product Hunt

  • Mudroomnew

    Mudroom runs coding agents in a VM and lets you review their changes like a pull request.

    #25 on Product Hunt

  • Trigger.dev: Webhooks for your AI agentsnew

    Trigger.dev adds webhooks so agents react to Slack, Supabase and 50 other services.

    #26 on Product Hunt

  • licensealnew

    Catches incompatible dependency licenses before release, useful for Android apps mixing GPL and commercial code.

    #27 on Product Hunt

  • Our Silly Gamesnew

    Skills for building games with Claude plus free multiplayer hosting.

    #28 on Product Hunt

  • Boltbarnew

    Spotlight-style AI chat for the Mac.

    #29 on Product Hunt

  • Agent Handshake by Fininew

    Agent-to-agent customer support that stands by its answer.

    #30 on Product Hunt

  • Trellisnew

    Trellis, a native macOS workspace for developers that is more than a terminal.

    #31 on Product Hunt

  • Notra

    A tool for ranking in AI search answers, the new SEO.

    #32 on Product Hunt

  • SavorApp_AI

    Captures notes and intentions and turns them into next steps.

    #33 on Product Hunt

  • Orbit

    Makes you earn each app opening to curb doomscrolling.

    #34 on Product Hunt

  • Nuvio 5.0

    Nuvio 5.0 hands financial admin to an agent.

    #35 on Product Hunt

  • fswitcher

    A replacement for Command-Tab on the Mac.

    #36 on Product Hunt

  • iPad mini 8

    Apple's iPad mini 8 with OLED, A20 Pro and IP67.

    #37 on Product Hunt

  • Paddock

    Paddock sells computers and sandboxes to agents as the user.

    #38 on Product Hunt

  • NINX

    Private on-device AI with connected tools and creative workflows.

    #39 on Product Hunt

  • ContextGem 1.0

    ContextGem 1.0 pairs decision models with LLMs to extract data from documents.

    #40 on Product Hunt

  • GPT-6.1 Terra

    GPT-6.1 Terra, OpenAI's cheaper computer-use model, still on Product Hunt.

    #41 on Product Hunt

  • Coachare

    A coaching app with a generic pitch and no stated method.

    #42 on Product Hunt

  • OpenRadView

    A free, open-source viewer for your own CT, MRI and X-ray scans.

    #43 on Product Hunt

  • Off Grid AI

    Off Grid AI, private AI on your phone and computer.

    #44 on Product Hunt

  • PulseDev

    Click any UI bug on localhost and your agent fixes it.

    #45 on Product Hunt

  • TheySaid 4.0

    Simulates how users will react to a design inside the build tool.

    #46 on Product Hunt

  • Revix AI

    Revix AI reviews code against a repo's own rules.

    #47 on Product Hunt

  • Overmind

    An open-source platform for continuous learning in AI systems.

    #48 on Product Hunt

  • NuPlay

    A platform for running a company's AI transformation, light on specifics.

    #49 on Product Hunt

  • Arcade Controller Pro

    A fight-game controller with buttons instead of a stick and its own screen.

    #50 on Product Hunt

Show HN

6

HN front page

30
Show 22 more

YC Launches

20
Show 12 more

HF trending models

19
Show 11 more

HF trending spaces

16
Show 8 more

HF daily papers

20
Show 12 more

Lobsters

25
Show 17 more

Dev.to top

20
Show 12 more

Techmeme

15
Show 7 more

About this issue

396 items from 77 sources, 10 of them seen in several sources. 3 sources failed to fetch.

Source health

77Which sources were fetched today, and how many items each returned.
SourceSectionItemsStatus
GitHub TrendingMain14Fetched
Product HuntMain50Fetched
Show HNMain6Fetched
HN front pageMain30Fetched
YC LaunchesMain20Fetched
HF trending modelsMain20Fetched
HF trending spacesMain20Fetched
HF daily papersMain20Fetched
LobstersMain25Fetched
Dev.to topMain20Fetched
TechmemeMain15Fetched
AndroidX releasesAndroid · Official & releases48Fetched
Android Developers BlogAndroid · Official & releases5Fetched
Android Studio releasesAndroid · Official & releases6Fetched
Kotlin blogAndroid · Official & releases2Fetched
Kotlin releasesAndroid · Official & releases0Fetched
Compose Multiplatform releasesAndroid · Official & releases1Fetched
Android Developers YouTubeAndroid · Official & releases0Failed
Kotlin on YouTube (JetBrains)Android · Official & releases0Failed
Google Developers BlogAndroid · Official & releases1Fetched
AndroidX merged API changesAndroid · AOSP / AndroidX insider24Fetched
Dove Letter (skydoves)Android · Experts5Fetched
skydoves on MediumAndroid · Experts0Fetched
Jake WhartonAndroid · Experts0Fetched
Zac SweersAndroid · Experts0Fetched
Chris BanesAndroid · Experts2Fetched
Ryan HarterAndroid · Experts0Fetched
Talking KotlinAndroid · Experts0Fetched
Joe BirchAndroid · Experts2Fetched
CommonsWare (Mark Murphy)Android · Experts3Fetched
TechYourChance (Vasiliy Zukanov)Android · Experts1Fetched
Kt. Academy (Marcin Moskała)Android · Experts2Fetched
Touchlab (Kotlin Multiplatform)Android · Experts2Fetched
Landscapist releasesAndroid · Library releases0Fetched
Balloon releasesAndroid · Library releases0Fetched
Sandwich releasesAndroid · Library releases0Fetched
Cloudy releasesAndroid · Library releases0Fetched
Compose Stability Analyzer releasesAndroid · Library releases0Fetched
Haze releasesAndroid · Library releases0Fetched
Compose Unstyled releasesAndroid · Library releases1Fetched
Coil releasesAndroid · Library releases0Fetched
Circuit releasesAndroid · Library releases0Fetched
Metro releasesAndroid · Library releases1Fetched
Molecule releasesAndroid · Library releases0Fetched
Turbine releasesAndroid · Library releases0Fetched
Paparazzi releasesAndroid · Library releases0Fetched
SQLDelight releasesAndroid · Library releases0Fetched
OkHttp releasesAndroid · Library releases2Fetched
Retrofit releasesAndroid · Library releases0Fetched
LeakCanary releasesAndroid · Library releases0Fetched
Koin releasesAndroid · Library releases0Fetched
Dagger / Hilt releasesAndroid · Library releases0Fetched
KSP releasesAndroid · Library releases0Fetched
kotlinx.coroutines releasesAndroid · Library releases0Fetched
kotlinx.serialization releasesAndroid · Library releases1Fetched
Ktor releasesAndroid · Library releases0Fetched
Apollo Kotlin releasesAndroid · Library releases0Fetched
Kermit releasesAndroid · Library releases0Fetched
Store releasesAndroid · Library releases0Fetched
Decompose releasesAndroid · Library releases1Fetched
Orbit MVI releasesAndroid · Library releases0Fetched
Roborazzi releasesAndroid · Library releases1Fetched
detekt releasesAndroid · Library releases0Fetched
ktlint releasesAndroid · Library releases0Fetched
MockK releasesAndroid · Library releases0Fetched
Lottie releasesAndroid · Library releases0Fetched
Authority Insights (leaks & teardowns)Android · Leaks & ecosystem8Fetched
9to5Google · AndroidAndroid · Leaks & ecosystem8Fetched
Android AuthorityAndroid · Leaks & ecosystem6Fetched
Android FaithfulAndroid · Leaks & ecosystem1Fetched
Android WeeklyAndroid · Community15Fetched
GitHub Trending · KotlinAndroid · Community11Fetched
Lobsters · androidAndroid · Community0Fetched
HN · AndroidAndroid · Community11Fetched
ProAndroidDevAndroid · Community5Fetched
Kotlin WeeklyAndroid · Community1Fetched
Philipp Lackner (YouTube)Android · Community0Failed
1 of 5

Android Studio Rabbit 2 is stable and runs any coding agent

You no longer have to choose between your preferred agent and the IDE's own tools, because the agent you already pay for now works inside Android Studio.

Google released Android Studio Rabbit 2 to the stable channel at Android Dev Summit. Its main feature is Bring Your Own Agent. You can run Claude Agent, Codex, Junie, Google Antigravity or any of over 40 agents that speak the open Agent Client Protocol inside the IDE, with access to the emulator, builds and Logcat. It also adds a skills manager, agent-assisted LeakCanary fixes, AI change review and stable XR emulators.

Android Developers Blog · OfficialRead the source

1 of 4

Opera lets websites use the AI on your computer

If browsers become the AI runtime, any site can add AI features at no per-request cost, which changes who can afford to build them.

WebAI by Opera topped Product Hunt. It lets a website call AI models that run on the visitor's own computer, through the browser, instead of sending requests to a server. For the site owner that means no API bill and no user data leaving the machine. It only works in Opera for now, so reach depends on whether other browsers adopt a similar standard.

Product Hunt · New ideaRead the source