Android Feed

← The 5 stories that mattered

Everything collected ·

All 371 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 · 2 min read

  1. Android Studio Rabbit 2 is stable with your own coding agent built in

    You can now use the coding agent your team prefers inside the stable Android Studio, with the IDE's project context.

    Google released Android Studio Rabbit 2 in the stable channel, ready for production use. The release brings the coding agents developers already use into the IDE, and Google lists Bring Your Own agent support among its key features. It was announced alongside Android Dev Summit, where Google described tools for working with agents.

    Android Developers Blog · OfficialRead the sourceProduct notes

  2. Glance gets adaptive widgets that target several surfaces

    Widget developers may soon describe one Glance widget that adapts to the home screen and other surfaces such as the watch.

    Jetpack Glance is Android's toolkit for building widgets with Compose-style code. Two merged AndroidX changes add an adaptive widget manager and new types for surfaces, widget instances and adaptive templates. A Dev Summit session also covers moving Wear Tiles to Wear Widgets with Glance. Nothing is released yet, so this is a direction, not an API to use today.

    AndroidX merged API changes · AOSP insiderRead the source

  3. CameraX is building a low-light boost effect

    Camera apps may soon get brighter results in the dark by adding one CameraX effect instead of writing their own image processing.

    CameraX is the Jetpack library most apps use for the camera. Five merged AndroidX changes add a low-light boost effect, with a session manager, a frame router and an OpenGL rendering pipeline on the graphics chip. That suggests a software boost that brightens dark preview and video, which could work on phones without a hardware low-light mode. It is not released yet.

    AndroidX merged API changes · AOSP insiderRead the source

  4. LiteRT in Play services gets faster GPU inference

    Apps with on-device AI features can get faster inference and a smaller download by switching to the Play services runtime.

    LiteRT is Google's runtime for running machine learning models on the device, the successor to TensorFlow Lite. When delivered through Google Play services, it now offers GPU inference powered by ML Drift, which Google says is about 1.4 times faster, plus a new CompiledModel API. Using the Play services version also keeps the runtime out of your app, so apps get smaller.

    Google Developers Blog · OfficialRead the sourceProduct notes

  5. Navigation 3 adds type-safe deep links

    If your app is modular and uses deep links, Navigation 3 can catch broken routes before users hit them.

    Navigation 3 is the newest Jetpack library for moving between screens in Compose apps. A new Android Developers video explains its type-safe deep links, which check link routes at compile time. In apps split into many modules, deep links often break quietly when a screen moves. Type-safe routes turn those into build errors instead of runtime surprises.

    Android Developers YouTube · OfficialRead the source

Also worth knowing

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

Go deeper: every Android item, by lane

Official & releases

22Android Studio Rabbit 2 is the stable IDE to move to for built-in agent support, LiteRT in Play services is worth trying for faster on-device GPU inference, and the Media3, Appstate and Test Core alphas can wait.

AndroidX library releases

Alpha · experimental, APIs may change · 4
Show 10 more

AOSP / AndroidX insider

28Today's merges build a CameraX low-light boost effect with its own GPU pipeline, add adaptive Glance widgets that target several surfaces, and keep shaping Remote Compose's public API.
  • Test tooling: read device facts and host logsnew

    Add DeviceFactsReader and HostLogManager

    Classes for reading device details and collecting host logs point to a new test harness or reporter that records the setup behind each test run.

  • Backup: a checker for backup rules filesnew

    Add BackupRulesChecker XML and env loader

    A checker that loads an app's backup rules XML suggests a tool or test helper that validates what gets backed up, likely to catch mistakes in backup and data extraction rules.

  • Test reports: a record for each test's detailsnew

    Add TestDetailRecord model and detailPath key

    A per-test detail record with its own path, merged next to the device facts reader, points to structured test results that link each test to its device and logs.

  • Benchmark: memory cost of shared element transitionsnew

    Add shared element memory macrobenchmark

    A memory macrobenchmark for shared element transitions means the Compose team is tracking their memory use, likely to catch regressions as the API matures.

  • IDE plugin: clear old files before updatingnew

    Remove old IDE plugin files before installing a new version

    Removing stale plugin files on install fixes upgrades of a plugin AndroidX ships for the IDE, a sign it is updated often enough for leftovers to matter.

  • Paging: helpers for lazy lists from snapshotsnew

    Add ItemSnapshotList extensions for Lazy foundations

    Extensions on ItemSnapshotList, Paging's snapshot of loaded items, for Lazy foundations suggest simpler ways to show paged data in Compose lazy lists.

  • Remote Compose: components become publicnew

    Expose RemoteComponent and drop RemoteFloatContext

    Exposing RemoteComponent and dropping RemoteFloatContext reshapes Remote Compose's public API, part of the clean-up before a wider release.

  • Remote Compose: dependencies fixed for app buildsnew

    Expose compose:remote dependencies used in public APIs as api

    Declaring dependencies used in public APIs as api means apps that use Remote Compose get the types they need without adding them by hand, a step toward outside use.

Show 20 more
  • Glance widgets: a manager for adaptive widgetsnew

    Add GlanceAdaptiveWidgetManager

    A GlanceAdaptiveWidgetManager suggests widgets that adapt their layout to where they are shown, likely tied to the move from Wear Tiles to Wear Widgets.

  • Glance: one widget for several surfacesnew

    Add GlanceSurface, AppWidgetSurface, WidgetInstanceInfo, and GlanceAdaptiveTemplate

    Surface, instance info and adaptive template types let one Glance widget describe itself for the home screen and other surfaces, pointing to a shared widget model across phone and watch.

  • CameraX: setting up low-light boost sessionsnew

    Add LowLightBoostSessionCreator

    A session creator is the entry point apps or CameraX would use to switch low-light boost on for a camera session.

  • CameraX: groundwork for low-light boostnew

    Add LowLightBoost session building blocks

    Shared building blocks for low-light boost sessions mean the feature is being built as a full pipeline, not a single flag.

  • CameraX: routing frames through low-light boostnew

    Add LowLightBoostFrameRouter

    A frame router sends camera frames through the boost step before preview and capture, so the effect can apply to more than one output.

  • CameraX: low-light boost as an effectnew

    Add LowLightBoostEffect, surface processor and session manager

    An effect with its own surface processor and session manager points to low-light boost shipping as a CameraX effect apps can add, not only a device camera mode.

  • CameraX: GPU pipeline to brighten dark framesnew

    Add LowLightBoost OpenGL rendering pipeline

    An OpenGL rendering pipeline suggests a software low-light boost that runs on the GPU, which could work on phones without a hardware low-light mode.

  • Test input shortcut name removed from public API

    Remove public InputDataFactory typealias

    Dropping a public typealias is API tidying that usually happens before a library moves from alpha to beta.

  • Screenshot tests: rule can be extended

    Add subclass extension points to ScreenshotTestRule

    Extension points on ScreenshotTestRule suggest AndroidX wants teams to build their own screenshot testing setups on top of it.

  • Remote widgets: button groups get a springy press

    Add spring press animation to RemoteButtonGroup

    Spring press motion on RemoteButtonGroup suggests Remote Compose widgets are getting the expressive Material motion that in-app Compose has.

  • Android XR: a builder for video surfaces

    Add SurfaceEntity.Builder

    A builder for SurfaceEntity makes it easier to place video or rendered surfaces in XR space, a sign the XR SceneCore API is being polished for wider use.

  • Remote Compose: basic icon goes to API review

    API Review F1: RemoteBasicIcon (remote-foundation)

    A formal API review of remote-foundation components points to Remote Compose's foundation layer heading for a public release.

  • Remote Compose: spacer goes to API review

    API Review F3: RemoteSpacer (remote-foundation)

    Part of the batched remote-foundation review, so the basic layout pieces of Remote Compose are being locked down.

  • Remote Compose: basic image goes to API review

    API Review F2: RemoteBasicImage (remote-foundation)

    Images are among the first remote-foundation components reviewed, which suggests widgets and tiles are the first target.

  • Remote Compose: foundation APIs cleaned for review

    Clear remote-foundation APIs and KDocs for batched review

    Cleaning APIs and docs in one batch is the step before a formal review, so a remote-foundation alpha looks close.

  • Remote Compose: basic text goes to API review

    API Review F4: RemoteBasicText (remote-foundation)

    With text, image, icon and spacer all in review, Remote Compose has the minimum set for simple widgets.

  • Accessibility speed flag removed

    Remove isAccessibilityPerformanceEnabled flag

    Removing a feature flag usually means the faster accessibility path is now always on, or was dropped; check release notes before relying on either.

  • Widgets: pick old or new renderer by Android version

    Pick RemoteViews or RemoteCompose by API level on one home screen surface

    One surface choosing RemoteViews or Remote Compose by API level suggests a path where widgets use Remote Compose on new Android and fall back on older devices.

  • Multiplatform libraries drop 32-bit Apple Watch

    Remove watchosArm32 target

    Removing the watchosArm32 target follows Kotlin's deprecation of that target, so libraries that publish for Apple Watch only ship 64-bit builds.

  • Widget templates: layout type kept internal

    Move the archetype from TemplateRenderer to internal concept so it is not exposed in public api and kept internally.

    Hiding the archetype from the public API means template-based widgets will expose fewer knobs, keeping room to change layouts later.

Experts

18Read Chris Banes on when an agent task needs a plan, Kt. Academy on Flow bugs that look correct, and Touchlab on hardening Turbine and coroutine tests.
Show 10 more

Library releases

12Haze 2.1.0 is the update to take for stable Glass, KSP 2.3.13 adds context parameters to its API, Metro 1.5.2, OkHttp 5.5.1, Ktor 3.6.1, Apollo 4.4.4 and Roborazzi 1.77.0 are safe patches, and Decompose 3.6.0 is still beta.
  • Haze 2.1.0

    See CHANGELOG.md for details.

  • Haze 2.0.1new

    See CHANGELOG.md for details.

  • Compose Unstyled 2.10.2

    Patch Changes 649c471 : Make dialogs, modal bottom sheets, and modal drawers behave inside of a Host the same as without. Tab and Shift+Tab now stay inside the topmost modal, background controls receive no keystrokes, and closing the modal restores the previous focus. b59481b …

  • Metro 1.5.2new

    Fixes [FIR] Fix compat with Kotlin 2.5.20 dev builds around inherited member injection. [IR] Fix graph extensions creating separate instances of inherited class binding containers. Extensions now reuse the parent instance and can share set and map getters. [IR] Fix a…

  • Metro 1.5.1

    Fixes [IR/IC] Avoid recompiling unrelated files in the graph module when the constructor of an upstream @Contributes* class changes. [Gradle] Fix false hidden-dependency reports for contributions from associated compilations, such as main sources in tests. This particularly…

  • OkHttp 5.5.1

    Version 5.5.1

  • OkHttp 4.12.1

    Version 4.12.1

  • KSP 2.3.13

    What's Changed Add KSContextParameter and KSFunctionDeclaration.contextParameters to the API ( #3216 ) Improve diagnostics by returning last round dismissible errors when KSP fails ( #3196 ) Add opt-in processor classloader caching ( ksp.classloader.cache.processors=true ) to…

Show 4 more
  • Ktor 3.6.1

    Published 29 October 2026 Improvements KTOR-10005 Native accepted TCP sockets do not enable TCP_NODELAY, unlike JVM KTOR-9966 Netty poor performance on delay calls KTOR-4206 The "com.typesafe.config.ConfigException$Missing" exception leaks from the typesafe config library when…

  • Apollo Kotlin 4.4.4

    A patch version to fix Java codegen generating bad code when using custom scalars mapped to Long or Float. 👷‍♂️ All changes [FIX] codegen: fix Java codegen generating bad code when using custom scalars mapped to Long or Float. ( #7032 )

  • Decompose 3.6.0-beta02

    What's Changed Library Add linuxX64 and linuxArm64 targets by @arkivanov-bot in #1028 Fix wrong directions in AndroidPredictiveBackAnimatableV1 in RTL layout by @arkivanov in #1030 Fix wrong directions in AndroidPredictiveBackAnimatableV2 in RTL layout by @arkivanov in #1031 Add…

  • Roborazzi 1.77.0

    Bugfix: verifyRoborazzi* and compareRoborazzi* reuse the configuration cache With the configuration cache enabled, the second verifyRoborazzi* or compareRoborazzi* run after clean recalculated the task graph because "the file system entry 'build/outputs/roborazzi' has been…

Leaks & ecosystem

21Microsoft Authenticator on Android is moving its password backup to Google One, Wear OS 8 gets a system status bar, and Android Auto 18.1 keeps preparing a light mode and redesign.
Show 13 more

Community

35A malformed WAP Push SMS that crashes Android's phone service is still the most discussed Android bug, an AR geology app is a popular Show HN, and Momentic's QA agents test Android apps by using them.
Show 27 more
Start the Product & AI read4 stories, ranked by what you’d regret missing · 2 min read

  1. Report says Postman quietly installs AI plugins and shares API data

    Developer tools now hold API keys and private data, so adding AI features without clear consent costs trust quickly.

    The top story on Hacker News, with about 1,050 points and nearly 350 comments, reports that Postman, the popular tool for testing APIs, silently installs AI plugins and attempts to share all API data. The brief gives only the report's title, so details and Postman's response are not included here. The discussion centers on consent and on how much data developer tools send.

    HN front page · Worth a debateRead the sourceDiscussion

  2. Vercel's agent-desktop speeds up to about 3,800 stars in a day

    Giving agents their own computer is becoming standard tooling, and this is the open-source option developers are picking up.

    Vercel Labs' agent-desktop is a command-line tool that lets AI agents operate a computer, with clicks, typing and screens. It added about 3,800 GitHub stars today, up from about 2,300 yesterday, and now has about 13 thousand. Product Hunt also lists redu.cloud, machines sized for agents, and Peji, an AI chat that drives a real browser.

    GitHub Trending · Fastest risingRead the sourceProduct notes

  3. Pushback on the claim that a model rewrote another's weights

    Big AI claims now get tested by others within a day, so a result that can be reproduced cheaply matters more than a headline.

    Yesterday's top story said GPT-6 Astra could read and write a small Qwen model's weights to improve it. Today a post with about 900 Hacker News points calls that claim misleading. Another post, with about 500 points, shows that Qwen 3.8 27B running on a laptop can interpret model weights too, which suggests the ability is less special than claimed.

    HN front page · Worth a debateRead the sourceDiscussion

  4. Anthropic's FlashProve proves which model answered

    As more apps reach models through routers and gateways, proof of which model answered could become a standard check.

    Anthropic released FlashProve, an open-source prototype that cryptographically verifies that an AI response came from a given model. The problem it targets is a router or reseller quietly swapping in a different model than the one you paid for. It is a prototype, and the brief does not say how fast or widely it can run.

    Techmeme · New ideaRead the sourceProduct notes

Go deeper: themes, product notes, every source

Themes across sources

6Agent sandboxes · Verifying agent work · Agent supply-chain safety · Self-improving agents · On-device AI · Coding-agent control surfaces

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

Agent sandboxes

Rising · 7 items from 4 sources

Vercel Labs' agent-desktop, a command-line tool that lets AI agents use a computer, added about 3,800 GitHub stars in a day, the most on the list. Product Hunt lists redu.cloud for machines sized for agents and Peji, a chat with a real browser you control, YC's Simantic simulates firmware for agents, and three Hugging Face papers study how computer-use agents learn and recover from mistakes.

Why now. Agents now click, type and run programs for long stretches, so they need a computer of their own that can be watched, reset and thrown away.

vercel-labs/agent-desktopGitHub Trending

Verifying agent work

Rising · 8 items from 4 sources

Momentic's QA agents that test web, iOS and Android apps by using them are still the most discussed Android launch on Hacker News. Product Hunt adds kwant, where bots check each other's work, plus Testifly and Testera for AI testing, YC's Workers IO simulates production to find failures, and Google's Android Bench session asks whether agent-written Android code is correct.

Why now. Agents now produce code faster than people can review it, so checking the result has become the slow step.

Launch HN: Momentic (YC W24) – Agents that catch bugs by using your appHN · Android

Agent supply-chain safety

Steady · 4 items from 4 sources

The top Hacker News story, with about 1,050 points, reports that Postman silently installs AI plugins and tries to share API data. Anthropic released FlashProve, an open-source prototype that proves a response came from the model you asked for, and Product Hunt's AgentKit and YC's Routant both sit between agents and the apps they are allowed to touch.

Why now. Agents and AI plugins now get access to real accounts and data, so what is installed, which model answers and what an agent may reach all need checks.

Postman silently auto-installs AI plugins and attempts to share all API dataHN front page

Self-improving agents

Steady · 5 items from 4 sources

Yesterday's claim that GPT-6 Astra edited a small Qwen model's weights drew pushback: a post calling it misleading has about 900 Hacker News points, and another shows Qwen 3.8 27B on a laptop can interpret weights too. The original weight-space post is still on Lobsters, and a Hugging Face survey covers video agents that improve themselves.

Why now. Models can now act on other models directly, and the debate has moved from whether it is possible to how much of it is new.

The 'Astra can read and write weights to improve Qwen' post is misleadingHN front page

On-device AI

Rising · 8 items from 6 sources

Liquid AI's LFM3-1.7B is the top trending model on Hugging Face for another day, with its small omni model and a browser demo also trending. On GitHub, Colibri streams model experts from disk and hex does voice dictation locally, while Google's LiteRT adds faster GPU inference through Play services and a Dev Summit session runs Gemini Nano in a Formula E car.

Why now. Small models are now good enough for daily tasks, and phones and laptops have the memory and GPU support to run them.

LiquidAI/LFM3-1.7BHF trending models

Coding-agent control surfaces

Rising · 7 items from 2 sources

Claude-Code-Game-Studios, which arranges 49 agents like a game studio, and Matt Pocock's skills collection each added over a thousand stars, and Worktrunk keeps parallel agents in separate Git worktrees. Product Hunt lists nana, one terminal for every coding agent on Windows, asklog for answering agents from your phone, Wavefunction and Friday Work.

Why now. Developers now run several agents at once, so juggling, watching and instructing them is a product category of its own.

Donchitos/Claude-Code-Game-StudiosGitHub Trending

More on this

Fading: Agent memory layers (last seen 2026-11-03) · Community-curated catalogs (last seen 2026-11-02) · Open-weight frontier models (last seen 2026-11-01) · Anti-slop design skills (last seen 2026-11-01) · Chat share-link exploits (last seen 2026-11-01) · World models (last seen 2026-10-31) · Agent-ready Android (last seen 2026-10-29) · Android XR glasses (last seen 2026-10-29)

Product notes

15Who 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

vercel-labs/agent-desktop

repoGitHub Trending★ 3,791 today

What
agent-desktop from Vercel Labs, a command-line tool that lets AI agents use a computer.
For whom
Developers building agents that must operate desktop apps or browsers, not just call APIs.
Problem
Most software has no API, so agents need screenshots, clicks and typing, which teams wire up by hand.
Why now
Models became good enough at reading screens and choosing actions that computer use is practical.
Wedge
A small CLI that any agent can call, from a company developers already trust for tooling.
Money
Open source; it likely feeds Vercel's hosted agent and sandbox products.
Like
Anthropic computer use · OpenAI Operator · Browserbase · Peji

LessonGrowth that speeds up on day two, from about 2,300 to 3,800 stars, means people are using it, not just bookmarking it.

Anthropic releases FlashProve, an open-source prototype that cryptographically verifies an AI response came from a given model, not one that a router swapped in (Anthropic)

newsTechmeme

What
FlashProve, an open-source prototype from Anthropic that cryptographically verifies an AI response came from a given model.
For whom
Companies buying model access through routers or resellers who need proof of which model answered.
Problem
A router can quietly swap in a cheaper model, and the customer has no way to tell.
Why now
Many apps now reach models through routers and gateways, which creates a gap between the model paid for and the model served.
Wedge
A narrow check, which model produced this answer, released openly as a prototype.
Money
Open source; it supports trust in Anthropic's paid models.
Like
Signed software releases · TLS certificates · C2PA content credentials

LessonWhen a middleman can cut corners unseen, a cheap way to verify the product becomes part of the product.

Show HN: Kairos – Open-source risk and fraud decisioning platform

productShow HN, HN front page64 pts · 7 comments

What
Kairos, an open-source risk and fraud decisioning platform from Finic, pitched as an alternative to Sift and Sardine.
For whom
Fintechs and marketplaces that need real-time decisions on payments, sign-ups and transactions.
Problem
Fraud platforms are closed and priced per decision, so teams cannot see or change why a transaction was blocked.
Why now
Fraud teams want rules and models they can inspect, and agents make writing and testing rules cheaper.
Wedge
Open source, so a team can self-host and read every decision path before paying anyone.
Money
Likely a hosted or enterprise version on top of the open-source core.
Like
Sift · Sardine · Unit21 · Stripe Radar

LessonIn a market of black boxes, being inspectable is a feature buyers in regulated industries will pay for.

Show HN: HackerTV – HN's most-recommended Youtube videos as 24/7 TV channels

productShow HN, HN front page188 pts · 39 comments

What
HackerTV, which plays the YouTube videos Hacker News recommends most as round-the-clock TV channels.
For whom
Hacker News readers who want something worth watching without picking it.
Problem
Good talks and documentaries are buried in years of comments, and choosing what to watch takes effort.
Why now
It builds on HackerTube, a recent project that mined 19 years of comments for recommended videos.
Wedge
Turning a list into a TV channel, so the viewer does not have to choose.
Money
Side project, no model.
Like
HackerTube · Pluto TV · YouTube playlists · Hacker Times

LessonThe same data in a new format can be a new product: a list asks you to choose, a channel does not.

kwant

productProduct Hunt#18 on Product HuntHype flag

What
kwant, AI bots that check each other's work before saying they are done.
For whom
Teams that hand multi-step tasks to agents and get back work that looks finished but is not.
Problem
A single agent marks its own homework, so errors reach people who then redo the checking.
Why now
Agents are cheap enough that running a second one to review the first costs less than a person's time.
Wedge
Cross-checking built in, rather than a single assistant.
Money
Likely a subscription priced by tasks or seats.
Like
Momentic · CodeRabbit · LangGraph · CrewAI

Agents checking agents can share the same blind spots; the listing gives no measure of how many errors it catches.

LessonSell the step users dread, checking, not the step that is already cheap, doing.

AgentKit by Scalekit

productProduct Hunt#3 on Product Hunt

What
AgentKit by Scalekit, an MCP gateway that lets production agents act on a user's behalf.
For whom
Teams putting agents in front of customers that need to call other services with the user's permission.
Problem
Each agent tool needs its own sign-in and permission handling, which teams rebuild for every integration.
Why now
MCP became the common way to connect agents to tools, which makes a shared gateway for sign-in and access worth building.
Wedge
Auth and delegation for agents, a piece teams find hard to build themselves.
Money
Likely usage or seat pricing on top of Scalekit's auth business.
Like
Routant · Arcade · Composio · Auth0

LessonAn auth company moving into agents shows that permission, not intelligence, is the hard part of putting agents in production.

nana

productProduct Hunt#8 on Product Hunt

What
nana, one terminal for every coding agent, built for Windows.
For whom
Windows developers who run several coding agents and switch between their command-line tools.
Problem
Most agent tooling targets macOS and Linux first, so Windows users juggle separate windows and setups.
Why now
Coding agents now run from the terminal, and several are in daily use at once.
Wedge
Windows first, a platform most agent terminals treat as an afterthought.
Money
Likely a paid app or subscription.
Like
Warp · Windows Terminal · t3code · Conductor

LessonWhen a category forms on one platform, the neglected platform is an open lane.

Routant — Let AI agents into your apps, on your rules

productYC Launches8 votes

What
Routant, which lets AI agents into your apps on your rules, with access to only the apps and data you choose.
For whom
Companies that want employees' agents to use internal tools without broad access.
Problem
Giving an agent an account often means giving it everything that account can see.
Why now
Agents now act inside business apps, and the Postman report shows how easily data can leave without people noticing.
Wedge
Selective access, choosing exactly which apps and data an agent can reach.
Money
Likely enterprise subscriptions.
Like
AgentKit by Scalekit · Okta · Arcade · Cloudflare Access

LessonEvery new kind of user, here agents, needs its own version of least privilege.

OpenHack: Always on AI Security Engineer

productYC Launches29 votesHype flag

What
OpenHack, an always-on AI security engineer that finds, verifies, prioritizes and fixes vulnerabilities.
For whom
Engineering teams without enough security staff to review every change.
Problem
Security scanners report many findings, most of them noise, and someone still has to verify and fix them.
Why now
Models can now read code well enough to confirm a bug is exploitable and write the fix.
Wedge
Verification before reporting, so only real issues reach people.
Money
Likely a subscription per repository or team.
Like
Snyk · Semgrep · GitHub Advanced Security · XBOW

Fixing the most complex vulnerabilities automatically is a strong claim with no results in the listing.

LessonIn a noisy category, the product that filters before it alerts wins the user's trust.

Clean: Real-time company data for sales teams and their agents

productYC Launches40 votes

What
Clean, real-time company data for sales teams and their agents, showing what just changed at each account.
For whom
Salespeople and sales agents who need a reason to contact an account today.
Problem
Sales data goes stale quickly, and reps miss the moment a customer's situation changes.
Why now
Sales agents now act on data directly, so fresh signals are worth more than large static lists.
Wedge
Changes, not profiles: what is new at an account and what to do about it.
Money
Likely a data subscription per seat or per agent.
Like
ZoomInfo · Apollo · Clay · Common Room

LessonSelling to agents as well as people changes what data is valuable: fresh and actionable beats big.

anomalyco/hex

repoGitHub Trending★ 410 today

What
hex, local voice dictation for Mac, Windows and Linux.
For whom
People who dictate to their computer, including to coding agents, and do not want audio sent to a server.
Problem
Good dictation usually runs in the cloud, which adds delay, cost and privacy worries.
Why now
Small speech models now run fast enough on ordinary laptops for live dictation.
Wedge
Local and cross-platform, where most rivals are Mac-only or cloud-based.
Money
Open source, no model yet.
Like
Wispr Flow · HoldTalk · MacWhisper · Superwhisper

LessonWhen a model shrinks enough to run locally, an open-source local version of a paid cloud product follows fast.

Stringman, an $800 open-source room scale gripper robot

discussionHN front page401 pts · 148 comments

What
Stringman, an open-source room-scale gripper robot that costs about 800 dollars.
For whom
Hobbyists, researchers and tinkerers who want a robot that can reach across a room.
Problem
Robot arms with useful reach are expensive and cover only a small area.
Why now
Cheap motors and open designs, plus robot-control models, make home robotics affordable to try.
Wedge
A low price and room-scale reach, likely from a cable-driven design instead of a long arm.
Money
Open source; possibly kits or parts sales.
Like
SO-100 arm · Hello Robot Stretch · Skycam cable systems · LeRobot

LessonChanging the shape of the problem, reach across a room instead of a precise arm, can cut the cost by an order of magnitude.

Your favorite agents are now integrated in Android Studio

articleAndroid Developers BlogAndroid

What
Android Studio Rabbit 2, now stable, which integrates the coding agents developers already use, with Bring Your Own agent support among its key features.
For whom
Android developers who use coding agents and want them inside the IDE rather than in a separate terminal.
Problem
Developers switch between Android Studio and outside agent tools, which lose the IDE's project and build context.
Why now
Teams now pick their own coding agent, so an IDE that only offers its own assistant gets bypassed.
Wedge
Open the IDE to other agents instead of competing with them.
Money
Free IDE that keeps developers in Google's tools.
Like
Cursor · JetBrains AI Assistant · VS Code agent mode · Claude Code

LessonWhen users already chose a tool you cannot beat, make your product the best place to use it.

Smaller apps and faster GPU inference with LiteRT in Google Play services

articleGoogle Developers BlogAndroid

What
LiteRT in Google Play services now offers GPU inference powered by ML Drift, about 1.4 times faster, with a new CompiledModel API.
For whom
Android apps that run machine learning models on the device, such as camera, vision or text features.
Problem
Apps bundle their own ML runtime, which makes them bigger, and GPU support varies across phones.
Why now
More apps run models on the device, so a shared, faster runtime from Play services saves size and work.
Wedge
Ship the runtime through Play services so apps get smaller and faster without bundling it.
Money
Free platform service that keeps on-device AI in Google's stack.
Like
TensorFlow Lite · ONNX Runtime Mobile · ExecuTorch · MediaPipe

LessonMoving a heavy dependency into the platform makes adoption easier, because the cost falls on the platform, not the app.

Show HN: Point your camera at a landscape to learn about its geology

discussionHN · AndroidAndroid261 pts · 55 comments

What
An app that paints geologic maps over landscapes in augmented reality, so you can point your camera at a view and learn its rocks.
For whom
Hikers, travelers and students curious about the land around them.
Problem
Geologic maps exist but are hard to read and hard to match to what you see in front of you.
Why now
Phone AR and location are accurate enough to line up map data with a real view.
Wedge
One fun use, identifying rocks in a view, built on public geologic data.
Money
Side project; possibly a paid app later.
Like
PeakFinder · Rockd · Google Lens · Sky Guide

LessonOverlaying existing public data on the camera view can turn a dry dataset into something people enjoy.

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

13
  • thedotmack/claude-mem

    claude-mem, automatic memory across agent sessions, adds about 1,100 stars and passes 112 thousand.

    ★ 1,139 today

  • vercel-labs/agent-desktopcard

    Computer use CLI for AI agents

    ★ 3,791 today

  • max-sixty/worktrunk

    Worktrunk manages Git worktrees so several coding agents can work on one repo in parallel.

    ★ 438 today

  • JustVugg/colibri

    Colibri runs large mixture-of-experts models on ordinary hardware by streaming cold experts from disk.

    ★ 802 today

  • mattpocock/skills

    Matt Pocock's agent skills collection adds about 1,200 stars and passes 310 thousand.

    ★ 1,164 today

  • p-e-w/heretic

    Training-free language model modification: Remove censorship, suppress slop, reduce sycophancy, and more

    ★ 701 today

  • spotify/portal-ai-pluginsnew

    Spotify publishes a repository of AI plugins for Portal, with no description yet.

    ★ 91 today

  • floci-io/flocinew

    Floci, a free local emulator for AWS services, as an alternative to LocalStack.

    ★ 482 today

Show 5 more
  • Usagi-org/ai-goofish-monitornew

    A Playwright and AI tool that monitors listings on Xianyu, China's second-hand marketplace.

    ★ 228 today

  • anomalyco/hexnewcard

    Local voice dictation for mac, windows and linux

    ★ 410 today

  • D4Vinci/Scrapling

    Scrapling, a web scraping framework from single requests to full crawls, adds about 1,700 stars.

    ★ 1,658 today

  • shiyu-coder/Kronos

    Kronos, a foundation model for financial market data, still trending.

    ★ 503 today

  • Donchitos/Claude-Code-Game-Studios

    Turns Claude Code into a game studio with 49 agents and 72 skills, adding about 1,200 stars.

    ★ 1,182 today

Product Hunt

50
  • NEXUPIRAnew

    NEXUPIRA, the top Product Hunt launch, pitches AI teammates that hand back finished work.

    #1 on Product Hunt

  • Teamily AInew

    Teamily AI puts your personal agent, agent teams and people in one messenger.

    #2 on Product Hunt

  • AgentKit by Scalekitnewcard

    MCP gateway for production agents that acts on your behalf

    #3 on Product Hunt

  • LifeValnew

    LifeVal, a benchmark of everyday life tasks for AI, listed as OpenAI's.

    #4 on Product Hunt

  • Akemanew

    Akema inspects Kafka messages read-only, so you can debug without risk of changing data.

    #5 on Product Hunt

  • Synter Canvasnew

    An AI agent that runs ad campaigns across more than 30 ad platforms.

    #6 on Product Hunt

  • Beamfynew

    Beamfy, a live AI vision assistant for blind people in 74 languages.

    #7 on Product Hunt

  • nananewcard

    One terminal for every coding agent, on Windows

    #8 on Product Hunt

Show 42 more
  • SUN Web & APInew

    Turns prompts and PDFs into audio, on the web or through an API.

    #9 on Product Hunt

  • Scrawlnew

    Scrawl lets you show your AI agent the problem by drawing a visual frame.

    #10 on Product Hunt

  • Wavefunctionnew

    Wavefunction, a minimal open-source coding agent for the Mac.

    #11 on Product Hunt

  • One Roofnew

    One Roof, a way to build with agents where the person stays in charge.

    #12 on Product Hunt

  • Brinknew

    Brink turns the edges of a MacBook trackpad into controls.

    #13 on Product Hunt

  • IQnew

    IQ, a dashboard of how a team uses AI.

    #14 on Product Hunt

  • MachineMailnew

    MachineMail gives email inboxes to both people and bots.

    #15 on Product Hunt

  • Testeranew

    Testera, AI-driven API testing.

    #16 on Product Hunt

  • EaseIn Healthnew

    An app to track GLP-1 doses, symptoms and daily habits together.

    #17 on Product Hunt

  • kwantnewcard

    AI bots that check each other's work before saying "done"

    #18 on Product Hunt

  • Hookdeck Inspectnew

    Hookdeck Inspect, a free webhook tester for you and your agent.

    #19 on Product Hunt

  • Flockbaynew

    Flockbay, a free AI game maker built on the Godot engine.

    #20 on Product Hunt

  • Beaquaro Studionew

    Write once and post to every social network, with a link-in-bio page.

    #21 on Product Hunt

  • MagicaLCorenew

    Trains six kinds of machine learning models on an iPad.

    #22 on Product Hunt

  • asklognew

    asklog gives coding agents a boss you can answer from your phone.

    #23 on Product Hunt

  • People Search by Context.devnew

    An API to find people by role, company and location.

    #24 on Product Hunt

  • Corvonew

    Corvo, a mailbox that sorts itself as mail arrives.

    #25 on Product Hunt

  • Pejinew

    Peji, an AI chat that drives a real browser you can take over.

    #26 on Product Hunt

  • BRCKS Chatnew

    A chat app for construction sites that keeps every message and photo findable.

    #27 on Product Hunt

  • LidCommitnew

    LidCommit makes a Git commit automatically when you close your laptop.

    #28 on Product Hunt

  • Calendarlynew

    Calendar and tasks on the iPhone Lock Screen.

    #29 on Product Hunt

  • Infino AInew

    Search, vectors and SQL over one Parquet copy of data, built for agents.

    #30 on Product Hunt

  • Testiflynew

    Testifly claims autonomous AI QA with no prompting and no upkeep.

    #31 on Product Hunt

  • Videconew

    Turns one video into a content campaign with AI.

    #32 on Product Hunt

  • PowerShipsnew

    A Battleship-style game with special powers.

    #33 on Product Hunt

  • HoldTalknew

    Hold the fn key, speak, and text appears in any Mac app.

    #34 on Product Hunt

  • VinSagenew

    A free used-car check in Chrome's side panel for the US and Canada.

    #35 on Product Hunt

  • Remote Gazenew

    Use your Mac from an iPad or iPhone without a relay server.

    #36 on Product Hunt

  • Prismnew

    Prism pitches itself as Cursor for hardware engineering, also a YC launch.

    #37 on Product Hunt

  • Oreja de Van Goghnew

    Pauses Apple Music while ChatGPT is speaking.

    #38 on Product Hunt

  • Display 1.0

    A shared workspace for creating documents and apps with any agent.

    #39 on Product Hunt

  • Help

    An app for asking for and giving targeted help in the open.

    #40 on Product Hunt

  • redu.cloud

    redu.cloud rents machines sized for agent workloads.

    #41 on Product Hunt

  • Xmind 2027

    Xmind 2027 turns its mind maps into an AI thinking partner.

    #42 on Product Hunt

  • Self-Growing Websites by Greta.sh

    Websites that run their own search and AI-answer optimization.

    #43 on Product Hunt

  • Friday Work

    Keeps a task running on another tool when Claude hits its usage limit.

    #44 on Product Hunt

  • Cedric AI for Meetings

    Cedric claims to present your slides and attend meetings for you.

    #45 on Product Hunt

  • diff-tour

    diff-tour teaches you to read AI-written code like a senior engineer.

    #46 on Product Hunt

  • Hera

    Hera, an AI creative canvas.

    #47 on Product Hunt

  • Hyperresearch

    Hyperresearch offers agentic search that checks its own findings.

    #48 on Product Hunt

  • beacon.sh

    beacon.sh, open-source memory for coding agents that it says improves itself.

    #49 on Product Hunt

  • Clayzo

    Create animations and motion design by asking your coding agent.

    #50 on Product Hunt

Show HN

11
Show 3 more

HN front page

30
Show 22 more

YC Launches

20
Show 12 more

HF trending models

19
Show 11 more

HF trending spaces

17
Show 9 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

371 items from 77 sources, 9 of them seen in several sources. 2 sources failed to fetch.

Source health

77Which sources were fetched today, and how many items each returned.
SourceSectionItemsStatus
GitHub TrendingMain13Fetched
Product HuntMain50Fetched
Show HNMain11Fetched
HN front pageMain30Fetched
YC LaunchesMain20Fetched
HF trending modelsMain20Fetched
HF trending spacesMain20Fetched
HF daily papersMain20Fetched
LobstersMain25Fetched
Dev.to topMain20Fetched
TechmemeMain15Fetched
AndroidX releasesAndroid · Official & releases13Fetched
Android Developers BlogAndroid · Official & releases4Fetched
Android Studio releasesAndroid · Official & releases5Fetched
Kotlin blogAndroid · Official & releases1Fetched
Kotlin releasesAndroid · Official & releases0Fetched
Compose Multiplatform releasesAndroid · Official & releases1Fetched
Android Developers YouTubeAndroid · Official & releases6Fetched
Kotlin on YouTube (JetBrains)Android · Official & releases0Failed
Google Developers BlogAndroid · Official & releases1Fetched
AndroidX merged API changesAndroid · AOSP / AndroidX insider28Fetched
Dove Letter (skydoves)Android · Experts5Fetched
skydoves on MediumAndroid · Experts0Fetched
Jake WhartonAndroid · Experts0Fetched
Zac SweersAndroid · Experts0Fetched
Chris BanesAndroid · Experts3Fetched
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 releases2Fetched
Compose Unstyled releasesAndroid · Library releases1Fetched
Coil releasesAndroid · Library releases0Fetched
Circuit releasesAndroid · Library releases0Fetched
Metro releasesAndroid · Library releases2Fetched
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 releases1Fetched
kotlinx.coroutines releasesAndroid · Library releases0Fetched
kotlinx.serialization releasesAndroid · Library releases0Fetched
Ktor releasesAndroid · Library releases1Fetched
Apollo Kotlin releasesAndroid · Library releases1Fetched
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 & ecosystem7Fetched
9to5Google · AndroidAndroid · Leaks & ecosystem8Fetched
Android AuthorityAndroid · Leaks & ecosystem6Fetched
Android FaithfulAndroid · Leaks & ecosystem1Fetched
Android WeeklyAndroid · Community15Fetched
GitHub Trending · KotlinAndroid · Community13Fetched
Lobsters · androidAndroid · Community1Fetched
HN · AndroidAndroid · Community6Fetched
ProAndroidDevAndroid · Community2Fetched
Kotlin WeeklyAndroid · Community1Fetched
Philipp Lackner (YouTube)Android · Community0Failed
1 of 5

Android Studio Rabbit 2 is stable with your own coding agent built in

You can now use the coding agent your team prefers inside the stable Android Studio, with the IDE's project context.

Google released Android Studio Rabbit 2 in the stable channel, ready for production use. The release brings the coding agents developers already use into the IDE, and Google lists Bring Your Own agent support among its key features. It was announced alongside Android Dev Summit, where Google described tools for working with agents.

Android Developers Blog · OfficialRead the source

1 of 4

Report says Postman quietly installs AI plugins and shares API data

Developer tools now hold API keys and private data, so adding AI features without clear consent costs trust quickly.

The top story on Hacker News, with about 1,050 points and nearly 350 comments, reports that Postman, the popular tool for testing APIs, silently installs AI plugins and attempts to share all API data. The brief gives only the report's title, so details and Postman's response are not included here. The discussion centers on consent and on how much data developer tools send.

HN front page · Worth a debateRead the sourceDiscussion