Android Feed

← The 5 stories that mattered

Everything collected ·

All 390 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. Google opens its developer docs to AI coding assistants

    Your coding agent can now check current Android docs before it writes code, which should cut outdated API suggestions.

    Google launched the Developer Knowledge API and an MCP server in public preview. MCP is the common way coding agents call outside tools. Together they let assistants search and fetch current official documentation for Android, Firebase, Google Cloud and other Google products, in a format machines can read, instead of guessing from what they learned during training.

    Google Developers Blog · OfficialRead the source

  2. Compose Material3 1.5 reaches beta

    If you use carousels, sliders or custom shapes, test the beta now so the stable update does not surprise you.

    Material3 for Jetpack Compose, the library behind Material Design components, moved version 1.5 to beta. The changes are mostly API clean-up before stable: a carousel parameter is renamed, the polygon shape APIs are reworked, and some older slider and app bar overloads are restored so code written for version 1.4 keeps working.

    AndroidX releases · OfficialRead the source

  3. JetBrains suggests Kotlin Toolchain for new multiplatform apps

    Starting a new shared Kotlin app may soon need far less build configuration than today.

    JetBrains released version 0.13 of the Kotlin Toolchain, a single tool that builds JVM, Android, iOS, multiplatform and server apps from a short declarative configuration. With this release, JetBrains recommends it for most new Kotlin Multiplatform projects, as a simpler starting point than a hand-written Gradle setup.

    Kotlin blog · OfficialRead the source

  4. Report says Google links your phone number to every signed-in account

    It is a reminder to check default account settings on test and personal devices that hold several Google accounts.

    Android Authority reports that Android verifies a phone number and links it to every Google account signed in on the device, through a setting that is on by default. Users noticed after receiving verification emails. The setting can be turned off in Google settings under privacy and security, but numbers already linked must be removed from each account by hand. Google had not commented.

    HN · Android · CommunityRead the sourceDiscussion

  5. Compose gets progressive blur, so do you still need Haze?

    You may be able to drop a dependency for simple blur effects and keep Haze only for frosted glass.

    Chris Banes, who maintains the Haze blur library, explains Compose's new progressive blur API. It blurs a composable's own content, which is different from Haze's backdrop effect that blurs what sits behind a surface. He also covers how both relate to Android's new native backdrop API, which helps decide when the built-in blur is enough.

    Chris Banes · ExpertRead 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

51The useful news is Google's Developer Knowledge API and MCP server, which give coding agents current Android docs; Material3 1.5 is in beta and Kotlin Toolchain 0.13 is now the suggested start for new KMP apps.

AndroidX library releases

Stable · safe to ship · 5

Beta · API frozen, try it before it ships · 15

Alpha · experimental, APIs may change · 16
Show 7 more
  • Start Your Next KMP App With Kotlin Toolchain 0.13

    The Kotlin Toolchain is a unified entry point into Kotlin. It lets you build JVM, Android, iOS, multiplatform, and server-side applications with a simple declarative configuration. With the release of v0.13, we now recommend the Kotlin Toolchain for most new Kotlin Multiplatform…

  • Discontinuing Swift Language IDE Support in the Kotlin Multiplatform Plugin

    The Kotlin Multiplatform plugin will stop offering Swift editing features from IntelliJ 2026.3.

  • 1.12.2

    Compose Multiplatform 1.12.2 fixes shadow flicker and an iOS text field crash.

  • buildHelpers_v0.1.2

    automatically created by TeamCity VCS labeling build feature

  • buildHelpers_v0.1.1

    automatically created by TeamCity VCS labeling build feature

  • 1.13.0-alpha02

    Changes since 1.13.0-alpha01 Highlights Web preloadFont , preloadImageBitmap , preloadImageVector are no longer experimental #5711 Migration Notes Multiple Platforms Temporarily re-exposed LocalSystemTheme (as @InternalComposeUiApi ) and deprecated it. It will be removed in a…

  • Supercharge your development with the Google Developer Knowledge API ecosystem

    Google is launching the Developer Knowledge API and MCP Server in public preview. This new toolset provides a canonical, machine-readable way for AI assistants and agentic platforms to search and retrieve up-to-date documentation across Firebase, Google Cloud, Android, and more…

AOSP / AndroidX insider

35Remote Compose is swapping Android drawing classes for its own types, Glance adaptive widgets are nearing a public API, and the photo picker is gaining albums, limits and custom styling.
  • Compose: trackpad pinch flag removednew

    Remove ComposeUiFlags.isTrackpadPinchReinterpretationEnabled

    The trackpad pinch behaviour has been settled, so the temporary feature flag goes away and the new behaviour becomes the default.

  • New people directory and interaction APIsnew

    Add PeopleDirectoryContentProvider and PeopleInteraction

    A content provider for people and their interactions suggests a shared contacts layer that apps or assistants can query.

  • Autofill content types get hintsnew

    Expose ContentType.contentHints

    Exposing content hints on ContentType points to better autofill and assistant understanding of text fields in Compose.

  • XR Compose owners updatednew

    Add users back to owners for XR Compose

    Housekeeping in the XR Compose team; no API change.

  • Compose host API can query semanticsnew

    Expose semantics and LayoutInfo queries under @ExperimentalComposeUiHostApi

    Semantics and layout queries on the experimental host API suggest tools or embedders reading Compose UI from outside, for example for testing or remote rendering.

  • Remote Compose: Wear widget profile removednew

    Remove RcPlatformProfiles.WEAR_WIDGETS and migrate callers

    Remote Compose is consolidating its platform profiles as Wear widgets move onto a shared path.

  • Glance adaptive heads toward public APInew

    Prepare glance-adaptive for public API

    Adaptive Glance widgets are close to a public release, likely for widgets that resize across phones, tablets and Wear.

  • Remote Compose docs filled innew

    RCA-108: Add missing KDocs and @param tags across Remote Compose APIs

    Adding missing KDocs is a usual step before an API leaves alpha.

Show 27 more
  • New Pixel devices added for testingnew

    Add pixel 10, 9a and 11 pro fold devices

    Pixel 10, 9a and 11 Pro Fold profiles in AndroidX tests point to foldable and new-device coverage.

  • Compose UI gets an embeddable hostnew

    Introduce ComposeUiHost, ComposeUiSharedResources, and @ExperimentalComposeUiHostApi

    ComposeUiHost and shared resources suggest running Compose UI inside other containers, such as remote or multi-window surfaces.

  • Benchmark for typing in text fieldsnew

    Add BasicTextField typing microbenchmark through InputConnection

    A typing microbenchmark means text input performance in BasicTextField is being measured, likely ahead of optimisation work.

  • Remote Compose: canvas renamednew

    Rename and migrate androidTest suites from RecordingCanvas to RemoteCanvas

    RecordingCanvas becomes RemoteCanvas, a naming cleanup before the API stabilises.

  • WebView startup takes a callback executornew

    Add startUpWebView overload accepting a callback Executor

    Apps can choose which thread receives WebView start-up callbacks, useful for keeping the main thread free.

  • Remote Compose: bitmap font drawing movednew

    Move drawBitmapFontTextRun APIs from RecordingCanvas to RemoteCanvas

    Font drawing APIs move to RemoteCanvas as part of the same API cleanup.

  • Remote Compose: brush tile modesnew

    Add RemoteTileMode and decouple RemoteBrush and RemoteImageBrush

    Remote Compose is getting its own drawing types instead of borrowing Android ones, so it can run on more platforms.

  • Remote Compose: image scaling typenew

    Add RemoteContentScale and decouple bitmap and color painters

    A remote-specific content scale type continues the move away from platform drawing classes.

  • Remote Compose: blend modesnew

    Add RemoteBlendMode and decouple RemoteColorFilter

    Its own blend mode type again points to Remote Compose becoming platform independent.

  • Remote Compose: font weight and stylenew

    Add RemoteFontWeight and RemoteFontStyle and decouple RemoteTextMeasure

    Text styling types are being defined inside Remote Compose for consistent rendering on remote surfaces.

  • Glance widget actions can launch intentsnew

    Add PendingIntent and callback receiver for GlanceWidgetAction

    PendingIntent support for GlanceWidgetAction lets adaptive widgets open activities or send broadcasts.

  • Glance adaptive gets widget actionsnew

    Add GlanceWidgetAction to adaptive-core

    Click actions arrive in the adaptive widget core, another step toward a public adaptive Glance API.

  • Content capture shims removed

    Remove shim classes in content capture

    Compatibility shims are no longer needed, likely because the minimum supported API level moved up.

  • Photo picker UI can be customisednew

    Add support for UI Customization Params in PhotoPicker

    Apps will be able to adjust the embedded photo picker's look to fit their own UI.

  • Remote Compose: default font familynew

    Add restricted defaultFontFamily RemoteTypography(Typography) constructor

    A typography constructor with a default font family helps map Material typography onto remote surfaces.

  • Photo picker selection limitsnew

    Add selection constraints params in photopicker

    Apps can set constraints on what users may select, such as counts or types, in the photo picker.

  • Shared axis transition fades in on backnew

    Add fadeIn to HorizontalSharedAxis backwardTransform

    A small polish to Material motion so backward navigation looks smoother.

  • Photo picker: albums and search highlightsnew

    Add support for Highlight Search Results and Album APIs in PhotoPicker

    Album access and highlighted search results point to a richer embedded photo picker.

  • Vertical text: horizontal runs insidenew

    [Tate] Add restricted HorizontalText composable

    Tate means Japanese vertical writing; this adds short horizontal runs such as numbers inside vertical text.

  • Vertical text: backgrounds for horizontal runsnew

    [Tate] Add horizontal text background helpers

    More groundwork for proper Japanese vertical layout in Compose.

  • Vertical text: layout and cachingnew

    [Tate] Add HorizontalTextLayout and Compose cache

    A layout and cache for horizontal runs suggests vertical text support is being built for real use, not only a demo.

  • CameraX test template: front camera configsnew

    Add front CameraStreamConfigurationMap to HighEndDeviceTemplate

    Test device templates get front camera stream maps to simulate high-end phones more accurately.

  • CameraX test template: default stream configsnew

    Add default CameraStreamConfigurationMap to HighEndDeviceTemplate

    Default stream configurations make CameraX tests on simulated devices closer to real hardware.

  • CameraPipe: old image wrapper removed

    Remove camera-camera2-pipe ImageWrapper

    Cleanup in the newer CameraPipe stack as older abstractions are retired.

  • Border renderer becomes public

    Exposes BorderRenderer as public API

    Exposing BorderRenderer lets developers reuse Compose's border drawing in custom components.

  • Owners list cleaned up

    Remove folks from owners who are no longer working on it

    Housekeeping; no API change.

  • Compose 1.13 beta API files tidied

    Remove stale bcv/native/current.ignore entries after 1.13.0-beta01

    Removing stale API-check entries after 1.13.0-beta01 means the beta API surface is now locked in.

Experts

17Read Chris Banes on Compose's new progressive blur and whether you still need Haze, and Touchlab's warning about crashes caused by mismatched transitive dependencies in Gradle.
Show 9 more

Library releases

3SQLDelight 2.4.1 and Compose Unstyled 2.10.1 are safe patch updates; Decompose 3.6.0 is still in beta, so wait unless you need its new Linux targets.
  • Compose Unstyled 2.10.1

    Patch Changes 9658157 : Support dismissOnClickOutside for DrawerPresentation.InPlace

  • SQLDelight 2.4.1

    Added [Documentation] Additional Android SQLite transaction documentation ( #6381 by @griffio ) Changed [Gradle Plugin] Keep parsed .sq files in memory for the whole code generation task, so they are not parsed again after garbage collection. This can make code generation faster…

  • 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…

Leaks & ecosystem

14Android 17 QPR2 Beta 7.1 brings a performance fix for testers, and reports say Google auto-links phone numbers across signed-in accounts.
Show 6 more

Community

43Google's android/skills repo for coding agents keeps climbing, LineageOS 24.0 is out, and an AOSP thread complains that late source releases force developers to disassemble libraries.
Show 35 more
Start the Product & AI read4 stories, ranked by what you’d regret missing · 2 min read

  1. Anthropic publishes an official directory of Claude Code plugins

    Curated plugin lists will decide which agent add-ons most developers ever see.

    Anthropic released a GitHub repository listing Claude Code plugins it manages and checks for quality, and it was the fastest-rising project on GitHub's trending page. Cursor published its own plugin specification and official plugins the same day. Both companies are moving from scattered community add-ons to curated lists they control.

    GitHub Trending · Fastest risingRead the sourceProduct notes

  2. Claude traces the origin of NIST's elliptic curve seeds

    It is a concrete case of a model settling an old open question that experts had argued over for years.

    Anthropic says Claude worked out that the random-looking seeds behind NIST's standard elliptic curves, used widely in ECDSA signatures, came from a pseudorandom generator fed with timestamps. For decades some suspected a hidden backdoor in those seeds. Cryptographer Filippo Valsorda wrote about the result, and it drew long threads on Hacker News and Lobsters.

    HN front page · Worth a debateRead the sourceDiscussion

  3. PostgreSQL 19 is delayed after AI-assisted reviews find design flaws

    AI review is now finding real design problems in one of the most carefully reviewed open-source projects.

    The PostgreSQL team pushed the general release of version 19 to November after code reviews assisted by language models found architectural defects. According to The Register, core developers are reverting up to ten features. They credit better review tooling for catching the problems rather than blaming the original authors.

    Techmeme · Worth a debateRead the source

  4. XERJ indexes your files so agents stop reading them line by line

    Token cost is now a design constraint, and tools that cut what agents read are getting quick adoption.

    XERJ is an open-source tool that indexes code, documents, logs and PDFs with one command. AI agents can then search the index instead of running grep and reading long outputs. The project says this uses about forty times fewer tokens, and that the same index serves retrieval, security audits and agent memory. It was among GitHub's fastest risers.

    GitHub Trending · Fastest risingRead the sourceProduct notes

Go deeper: themes, product notes, every source

Themes across sources

6Coding-agent control surfaces · Agent harnesses · Agent-ready Android · Agent sandboxes · On-device AI · Open-weight frontier models

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

Coding-agent control surfaces

Steady · 7 items from 2 sources

Product Hunt filled up with apps whose whole job is to watch and steer several coding agents at once: Parallax, Harness by Autonomous, Atelier, Palomino, BOSS Console and Toad. They differ in where the agents run, but all of them treat the agent as a worker you supervise rather than a chat you type into.

Why now. Developers now run several agents in parallel on real tasks, so the bottleneck has moved from writing code to keeping track of what each agent is doing.

ParallaxProduct Hunt

More on this

Agent harnesses

Steady · 6 items from 4 sources

The setup around a model, its skills, memory, rules and hooks, keeps drawing the most GitHub stars: ECC, ruflo and superpowers all trend again. A YC company proposes an open standard for running whole harnesses as a product backend, and a new paper lets a harness rewrite its own prompts and workflows.

Why now. Models are good enough that the scaffolding around them now decides most of the result, and that scaffolding can be shared and versioned like code.

affaan-m/ECCGitHub Trending

More on this

Agent-ready Android

Rising · 4 items from 4 sources

Google is feeding coding agents official Android knowledge from several directions: a Developer Knowledge API and MCP server for docs, the android/skills repo of agent instructions, and the Gemini model post on the Android blog. The community adds tools such as a headless jadx server that lets agents read APKs.

Why now. Agents that guess from stale training data write outdated Android code, so platform owners now ship machine-readable sources directly to them.

Supercharge your development with the Google Developer Knowledge API ecosystemGoogle Developers Blog

More on this

Agent sandboxes

Steady · 4 items from 3 sources

Running agents somewhere safe and cheap is becoming its own market. Agent37 launched low-cost sandboxes on Hacker News, Scalebrowser and Pomerado give agents a real browser, and context-mode sandboxes noisy tool output so it does not flood the model's context.

Why now. Agents now run for hours and call many tools, which makes isolation and per-run cost matter as much as model quality.

Launch HN: Agent37 (YC F26) – Low-cost agent sandboxes for running agentsHN · Android

More on this

On-device AI

Rising · 6 items from 3 sources

Hugging Face's trending list is led by small and efficient models: Google's EmbeddingGemma 2, MiniCPM-V 4.7 in a 1B size and a mixture-of-experts size, and Liquid AI's LFM2.5. Dev.to posts measure Gemma 4 drafters on small GPUs and describe fine-tuning small models for on-device agents.

Why now. Small models have become accurate enough for search, vision and simple agent steps, so running them locally saves both cost and privacy trouble.

google/embeddinggemma-2HF trending models

Open-weight frontier models

Steady · 4 items from 3 sources

Meta's Muse Spark Open is on Product Hunt and Hugging Face, and Apple's ML research team shows it running with MLX on a Mac Studio. NVIDIA adds Nemotron 3.5 Super VL, an open multimodal model aimed at specialised agents.

Why now. Open weights at near-frontier quality let teams run strong models on their own hardware instead of renting them.

Apple ML Research: Muse Spark Open with MLX on Mac StudioHN front page

More on this

Fading: Small decision models (last seen 2026-10-16) · AI-driven reverse engineering (last seen 2026-10-16) · Agent memory layers (last seen 2026-10-15) · Verifying agent work (last seen 2026-10-15) · MCP-native apps (last seen 2026-10-13) · Self-improving agents (last seen 2026-10-12)

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

anthropics/claude-plugins-official

repoGitHub Trending★ 1,552 today

What
An official directory of Claude Code plugins, curated and managed by Anthropic, published as a GitHub repository.
For whom
Developers using Claude Code who want add-ons they can trust without vetting random repositories.
Problem
Plugins and skills are scattered across thousands of personal repos of uneven quality, so people copy setups from social posts and hope.
Why now
Plugins became a standard way to extend coding agents this year, and a large ecosystem needs a trusted default list.
Wedge
Being the official, vendor-run list that ships with the tool's own name on it.
Money
Free; it supports usage of Anthropic's paid models and Claude Code subscriptions.
Like
cursor/plugins · VS Code Marketplace · Homebrew core

LessonOnce an ecosystem grows, the platform owner's curated list becomes the main way people find extensions, so being listed matters more than being early.

xerj-org/xerj

repoGitHub Trending★ 1,648 todayHype flag

What
An open-source tool that indexes code, docs, logs and PDFs with one command so AI agents can search them instead of running grep and sed.
For whom
Teams whose agents waste tokens reading large repositories and document folders line by line.
Problem
Agents search with shell tools and paste long outputs into their context, which is slow and expensive.
Why now
Agent token bills became a real line item, so a search index that cuts reads is easy to justify.
Wedge
A single indexing command that also serves RAG, security audits and agent memory.
Money
Open source, no model yet.
Like
ripgrep · Sourcegraph · LlamaIndex

The 40x fewer tokens figure comes from the project itself and depends heavily on the workload.

LessonSelling a cost cut against a tool everyone already uses, here grep, makes the value easy to measure and explain.

mksglu/context-mode

repoGitHub Trending★ 449 todayHype flag

What
A context-window optimiser for coding agents that sandboxes tool output, keeps session memory and routes requests across many agent platforms through MCP and hooks.
For whom
Heavy coding-agent users who hit context limits during long sessions.
Problem
Large tool outputs fill the context, so the agent forgets earlier work or the session has to restart.
Why now
Hooks and MCP now exist in most agent tools, giving a standard place to intercept output before the model sees it.
Wedge
One install that works across 17 agent platforms instead of a plugin per tool.
Money
Open source, no model yet.
Like
Claude Code compaction · Aider repo map

The 98% reduction is the project's own figure for tool output, not for whole sessions.

LessonSitting between two systems, the agent and its tools, lets a small project improve many products without integrating with each one.

Parallax

productProduct Hunt#17 on Product Hunt

What
A desktop app that shows every coding agent you are running in one place.
For whom
Developers who run several agents across terminals and editors at the same time.
Problem
Parallel agents live in separate windows, so people lose track of which one is waiting, stuck or finished.
Why now
Running several agents in parallel only became normal this year.
Wedge
A single overview screen rather than a new agent.
Money
Not stated; likely a paid app or subscription.
Like
Harness by Autonomous · Orca · tmux

LessonWhen a new kind of work appears, the first useful product is often just a dashboard for it.

BOSS Console

productProduct Hunt#1 on Product HuntHype flag

What
A small operating layer for agents that run all the time, described as a tiny OS for always-on agents.
For whom
Builders who keep agents running in the background on a schedule or on events.
Problem
Long-running agents need process control, logs and restarts, which people currently glue together with scripts.
Why now
Agents moved from one-off chats to background jobs that need supervision.
Wedge
The operating-system framing for a single always-on agent before larger fleets.
Money
Not stated.
Like
Shrine · pm2 · systemd

The listing gives a framing, not concrete features or limits.

LessonBorrowing a familiar mental model, here an OS, can explain a new category in three words.

Nudgebox

productProduct Hunt#11 on Product Hunt

What
A service for sending push notifications to people without building your own app.
For whom
Small businesses, communities and creators who want phone notifications but have no mobile team.
Problem
Push notifications usually require shipping and maintaining an Android and iOS app, which most small senders cannot justify.
Why now
Email open rates keep falling and web push is limited on some platforms, so a shared notification app becomes attractive.
Wedge
Offering just the notification channel rather than a whole app builder.
Money
Likely paid per sender or per message volume.
Like
ntfy · Pushover · OneSignal

LessonFor Android developers: the feature people want from an app is sometimes only its notification, and that can be sold on its own.

Launch HN: Agent37 (YC F26) – Low-cost agent sandboxes for running agents

discussionHN · AndroidAndroid162 pts · 127 comments

What
Sandboxes for running AI agents, launched by a YC company that says it costs under a tenth of most sandbox providers.
For whom
Teams that need many isolated environments for agents and pay too much for each one.
Problem
Sandbox pricing built for short code runs gets expensive when agents keep environments alive for long tasks.
Why now
Agent workloads now need thousands of cheap, long-lived environments rather than a few short ones.
Wedge
Price: the founders say customers average 37 cents a month per sandbox.
Money
Usage-based pricing per sandbox.
Like
E2B · Modal · Daytona

LessonIn a crowded infrastructure category, a much lower price for the common case is still a real wedge.

PlaceCall by VOYGR – The phone is the last API

productYC Launches100 votes

What
An API that finds the right local businesses, calls them in parallel and returns answers, quotes and bookings.
For whom
Apps and agents that need information only available by phone, such as availability or prices.
Problem
Many businesses have no booking system or API, so someone has to call them one by one.
Why now
Voice models can now hold short business calls well enough to automate them.
Wedge
Treating the phone call as an API endpoint.
Money
Likely per call or per completed task.
Like
Google Duplex · Bland AI · Vapi

LessonWhere the world has no API, the product can be the API that wraps the old channel.

Proof of Human, PBC: The invisible Turing Test for the Internet

productYC Launches13 votes

What
An API that keeps checking whether a visitor is a human or a bot or AI agent, without a visible challenge.
For whom
Sites and apps that must treat human users and automated agents differently.
Problem
CAPTCHAs annoy people and are increasingly solved by agents, so they no longer separate the two.
Why now
Browser agents now act on the web on people's behalf at scale.
Wedge
Continuous, invisible checking rather than a one-time challenge.
Money
API pricing; it is set up as a public benefit corporation.
Like
Cloudflare Turnstile · reCAPTCHA · World ID

LessonEvery new kind of user, here AI agents, creates demand for tools that tell users apart.

google/embeddinggemma-2

modelHF trending modelstrending 939

What
The second version of Google's EmbeddingGemma, a small embedding model, now at the top of Hugging Face's trending models.
For whom
Developers building search, recommendations or RAG that should run on a phone or small server.
Problem
Good embeddings usually come from large hosted models, which adds cost, latency and privacy concerns.
Why now
Small embedding models have closed much of the quality gap with large ones.
Wedge
An open model from Google sized for on-device use.
Money
Free open weights that draw developers to Google's Gemma family and tools.
Like
EmbeddingGemma · nomic-embed · BGE

LessonFor Android developers: semantic search inside an app no longer needs a server if the embedding model fits on the device.

openbmb/MiniCPM-V-4.7-35B-A3B

modelHF trending modelstrending 689

What
MiniCPM-V 4.7, an open vision-language model from OpenBMB in a mixture-of-experts size with about 3 billion active parameters.
For whom
Developers who need image understanding on modest hardware.
Problem
Vision models strong enough for documents and screenshots are usually too heavy to run locally.
Why now
Mixture-of-experts designs keep quality while running only a small part of the model per step.
Wedge
A strong vision model that is cheap to run, with a 1B sibling for even smaller devices.
Money
Open weights; research lab, no direct model.
Like
Qwen-VL · Gemma vision · SmolVLM

LessonShipping a family of sizes lets one release serve servers and phones at once.

android/skills

repoGitHub Trending, GitHub Trending · KotlinAndroid★ 36 today

What
Google's android/skills repository: ready-made instructions that teach coding agents how to do Android tasks.
For whom
Android developers who use coding agents and want them to follow current Android practice.
Problem
Agents write Android code from old training data and miss newer APIs and patterns.
Why now
Skills files became a common format that several agent tools can read.
Wedge
Official guidance from the Android team, packaged in the format agents already load.
Money
Free; supports the Android platform and Google's AI tools.
Like
Google Developer Knowledge API · Context7

LessonPlatform owners now document for two readers, people and agents, and the agent version needs its own format.

orailnoor/DroidDesk

repoGitHub Trending · KotlinAndroid★ 111 today

What
A project that turns an Android phone into a Linux desktop using Termux, Termux X11, TUR and Proot, so you can run VS Code, Firefox, LibreOffice and Blender.
For whom
Tinkerers and students who want a full desktop from a phone and a monitor.
Problem
Setting up a desktop Linux on Android without root takes many manual steps across several tools.
Why now
Phones are powerful enough, and Android's own desktop mode work makes the idea feel practical.
Wedge
Packaging existing tools into one guided setup rather than writing a new system.
Money
Open source, no model.
Like
Samsung DeX · Andronix · UserLAnd

LessonBundling several hard-to-combine tools into one setup can be a product by itself.

1013503897/jadx-headless-mcp

repoGitHub Trending · KotlinAndroid★ 7 today

What
An MCP server for static analysis of Android APKs, built directly on jadx-core with no GUI needed.
For whom
Security researchers and Android developers who want an agent to inspect compiled apps.
Problem
Reverse engineering an APK means clicking through the jadx GUI by hand.
Why now
MCP gives agents a standard way to call tools such as a decompiler.
Wedge
Headless operation, so it works on servers and inside agent loops.
Money
Open source, no model.
Like
jadx · rea · apktool

LessonWrapping a trusted existing tool for agents is faster to build and easier to trust than a new one.

Show HN: rStation – A shared research lab for your AI agents

productShow HN50 pts · 19 commentsHype flag

What
rStation, presented on Show HN as a shared research lab where several AI agents work together.
For whom
People running multi-step research with more than one agent.
Problem
Separate agent sessions cannot see each other's findings, so work is repeated.
Why now
Agents now run long enough that sharing results between them matters.
Wedge
A shared workspace rather than a smarter single agent.
Money
Not stated.
Like
Notion · CrewAI · AutoGen

The launch post gives little detail on what the lab actually stores or how agents coordinate.

LessonShared state, not more intelligence, is often what multi-agent setups lack.

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
  • anthropics/claude-plugins-officialcard

    Official, Anthropic-managed directory of high quality Claude Code Plugins.

    ★ 1,552 today

  • K-Dense-AI/scientific-agent-skills

    A large shared library of science-specific agent skills, a sign that skills are becoming domain packs rather than generic prompts.

    ★ 1,158 today

  • affaan-m/ECC

    One of the most-starred agent harnesses keeps trending, showing demand for packaged skills, memory and safety around coding agents.

    ★ 1,057 today

  • HKUDS/CLI-Anythingnew

    Tries to give any desktop software a command line so agents can drive it without screen clicking.

    ★ 281 today

  • mksglu/context-modecard

    Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.

    ★ 449 today

  • ruvnet/ruflo

    A multi-agent swarm framework that keeps trending, though its claims are broad and hard to compare.

    ★ 672 today

  • android/skillscardin 2 sources

    ★ 36 today

  • obra/superpowers

    A skills framework and working method for coding agents that remains one of GitHub's most-starred agent projects.

    ★ 973 today

Show 5 more
  • cursor/pluginsnew

    Cursor publishes its plugin specification and official plugins, the same move Anthropic made the same day.

    ★ 278 today

  • debpalash/VoiceStudio

    A fully local alternative to ElevenLabs for voice cloning, dubbing and transcription that keeps rising.

    ★ 2,301 today

  • MadsLorentzen/ai-job-searchnew

    A local job-search toolkit built on Claude Code, showing agents moving into personal admin work.

    ★ 898 today

  • hmjz100/LinkSwiftnew

    一个基于 JavaScript 的网盘文件下载地址获取工具。基于【网盘直链下载助手】修改 ，支持 百度网盘 / 阿里云盘 / 中国移动云盘 / 天翼云盘 / 迅雷云盘 / 夸克网盘 / UC网盘 / 123云盘 八大网盘

    ★ 538 today

  • xerj-org/xerjcard

    XERJ is the new way for AI to search data. Its autoindex capability activates agents to know your data without the token waste of grep and sed. One command indexes code, docs, logs and PDFs for search, RAG, security audits and agent memory, using 40x fewer tokens than grep…

    ★ 1,648 today

Product Hunt

50
  • BOSS Consolenewcard

    A tiny OS for your always-on agents

    #1 on Product Hunt

  • Holmnew

    A self-hosted home page shared by everyone in a household.

    #2 on Product Hunt

  • Lovable Growth Toolsnew

    Lovable extends from building an app to growing it, moving app builders toward marketing work.

    #3 on Product Hunt

  • Calenterinew

    SaaS where a user can can book appointments to other users.

    #4 on Product Hunt

  • Scalebrowsernew

    Gives AI agents a browser with human-like capabilities, aimed at sites that block automation.

    #5 on Product Hunt

  • Ateliernew

    A workspace designed for coding agents rather than humans, another sign that agents get their own tools.

    #6 on Product Hunt

  • Sparrownew

    Self-hosted webhook delivery as one binary on Postgres, a small alternative to hosted webhook services.

    #7 on Product Hunt

  • Palominonew

    A fast workspace for running agents in parallel, competing with several similar launches today.

    #8 on Product Hunt

Show 42 more
  • Agent Vaultnew

    A credentials and access store for agents, as people worry about handing them real keys.

    #9 on Product Hunt

  • Tidy Designnew

    An open-source design tool pitched as an alternative to Figma and Paper.

    #10 on Product Hunt

  • Nudgeboxnewcard

    Send push notifications without building an app

    #11 on Product Hunt

  • Toadnew

    Lets you point an agent at any machine you own, turning spare computers into agent workers.

    #12 on Product Hunt

  • Mayblocnew

    Claims state-of-the-art speech-to-text for audio with several speakers, a hard case for meeting tools.

    #13 on Product Hunt

  • iKnowMyMacnew

    Explains why a Mac is slow with evidence and a suggested fix, a diagnostic tool for non-experts.

    #14 on Product Hunt

  • Heliconnew

    An open-source desktop app and VS Code panel for Muse Code, Meta's coding assistant.

    #15 on Product Hunt

  • Kalesh Judgenew

    Turn petty fights into a public trial

    #16 on Product Hunt

  • Parallaxnewcard

    Every coding agent you run, in one place.

    #17 on Product Hunt

  • Muse Spark Opennew

    Meta's open-weight Muse Spark model, presented as the most capable US open model so far.

    #18 on Product Hunt

  • Tesla Roadsternew

    The hypercar that flies

    #19 on Product Hunt

  • DOYTnew

    The Dynamic Island to-do app you don't have to open

    #20 on Product Hunt

  • Tetriz Harnessnew

    Turns what a team learns into shared agent setup, so each developer's agent improves together.

    #21 on Product Hunt

  • GBrain Toolsnew

    A password manager built for AI agents, part of the push to keep real credentials out of prompts.

    #22 on Product Hunt

  • Mermaidnew

    Tell your Mac what to do. From literally anywhere

    #23 on Product Hunt

  • Harness by Autonomousnew

    A command center for every coding agent on every machine, another entrant in agent management.

    #24 on Product Hunt

  • Jockey by TwelveLabsnew

    TwelveLabs makes a whole video library searchable and organised by its content.

    #25 on Product Hunt

  • FinderGitnew

    Every Git repo on your Mac: browse, commit, review, clean

    #26 on Product Hunt

  • Memory Chessnew

    Study a chess position, then rebuild it from memory

    #27 on Product Hunt

  • Taimnew

    Capture it. Taim files, schedules & syncs it with your AI.

    #28 on Product Hunt

  • Skissonew

    Edit a live UI inside your design system and hand off the code diff, linking design and code.

    #29 on Product Hunt

  • Fish Audio Drama 3new

    AI voice acting you can direct line by line rather than only generate.

    #30 on Product Hunt

  • Firefox Enterprisenew

    Firefox for enterprises that want control of their data

    #31 on Product Hunt

  • Neophytenew

    Watch an AI agent sign up and onboard to any product

    #32 on Product Hunt

  • Nemotron 3.5 Super VLnew

    NVIDIA's open multimodal model aimed at specialised agents.

    #33 on Product Hunt

  • Dottrnew

    Agent that drafts contracts, flags risks + gets them signed

    #34 on Product Hunt

  • Pomeradonew

    Turns any website into an API for agents, betting browser automation can be made reliable.

    #35 on Product Hunt

  • Grablo Spotnew

    Free AI NVR that finds any person or vehicle in seconds

    #36 on Product Hunt

  • NANI???new

    Your interpreter and your boss doesn't even know.

    #37 on Product Hunt

  • SignWithnew

    E-signatures without the subscription. Pay per doc.

    #38 on Product Hunt

  • Hintly for Google Meet™new

    Real-time hints for every term you miss on Google Meet

    #39 on Product Hunt

  • Shrine

    A home for your agents

    #40 on Product Hunt

  • siliXon

    Turn a plain English idea into a circuit board you can order

    #41 on Product Hunt

  • Audio Lip Sync Pro for Blender

    Generate high-quality facial animations, direct from audio.

    #42 on Product Hunt

  • Chatway MCP

    Lets ChatGPT and Claude answer support tickets through MCP, putting support desks inside chat assistants.

    #43 on Product Hunt

  • Audimate

    AI assistant that works right in the apps you already use

    #44 on Product Hunt

  • CmdRound

    Assess how engineers build with artificial intelligence

    #45 on Product Hunt

  • Kerno 1.0

    The AI Runtime Backend QA Engineer

    #46 on Product Hunt

  • Actually Asked

    A self-improving branded AI FAQ page

    #47 on Product Hunt

  • Superdraft

    Automatically identify opportunities to email your customers

    #48 on Product Hunt

  • Rushilo

    Edit video by talking to Claude Code or Codex, coding agents applied to media work.

    #49 on Product Hunt

  • CinemaLut

    Any frame's color, as a LUT for yours.

    #50 on Product Hunt

Show HN

4

HN front page

30
Show 22 more

YC Launches

20
Show 12 more

HF trending models

18
Show 10 more

HF trending spaces

16
Show 8 more

HF daily papers

20
Show 12 more
  • Mosaic: Multi-Session Agentic Reinforcement Learningnew

    Most existing agentic reinforcement learning methods assume single-session trajectories, where interaction proceeds within one continuously accumulated context. However, agentic techniques such as context compaction and subagent forking may produce trajectories spanning multiple…

    48 upvotes

  • Can World and Action Modeling Benefit Each Other under Unified Causal Training?new

    World and action modeling are closely related, yet their reciprocal benefits under unified causal training remain unclear. We introduce OneWM, a simple framework for unified causal training that treats actions as both conditions for world prediction and targets for policy…

    39 upvotes

  • Behavioral Compositionality of Sparse Autoencoder Representationsnew

    Sparse autoencoders (SAEs) represent model activations with sparse, linearly decoded features, but the downstream effects of those features need not compose additively. We introduce SAE Additivity Residuals (SAR), which measure the behavioral variation left unexplained above a…

    35 upvotes

  • WorldCast: Exploring Intelligent Visual Generation in World State Predictionnew

    Humans interpret multimodal observations of the world and forecast how it will evolve. Recent multimodal generative systems have demonstrated similar capabilities: they can integrate multimodal context and synthesize visual outputs that depict changes in the world. To…

    29 upvotes

  • OmniVerifier-C1: Decomposed Graded Verifier for Rule-Governed Generative World Modelingnew

    Generative visual verifiers play a crucial role in advancing multimodal intelligence. To enhance the robustness of visual verifiers in complex multimodal scenarios involving compositional requirements, we move beyond conventional binary verifiers that produce only True/False…

    27 upvotes

  • Data World: Scaling Data Environments from Scratchnew

    Data agents use code to process heterogeneous files and derive analytical insights. Scaling their training faces costly environment curation, privacy and intellectual-property restrictions, and difficulty verifying open-ended findings without known generative truth. Our core…

    25 upvotes

  • MLExpert: Towards a Generalist Agentic Model for Autonomous Machine Learning Engineeringnew

    Large language models (LLMs) are expanding from assisting scientific research and engineering to autonomously developing and improving AI systems, advancing AI for AI (AI4AI) as a pathway toward recursive self-improvement (RSI). Workflow-driven machine learning engineering (MLE)…

    24 upvotes

  • Each Reward to Its Relevant Tokens: Reward-Specific Token-Level Credit Assignment for Multi-Reward RLnew

    Language models are increasingly trained against several verifiable rewards at once, such as the correctness and the format of a tool call, or the individual constraints of a complex instruction. Reward-decoupled normalization (GDPO) normalizes each reward separately so that…

    20 upvotes

  • Trajectory-Level Speculation with Semantic Verification for Tool-Using Agentsnew

    Large language model agents solve tasks through long sequences of tool calls, and the prevailing design has a single strong model generate every step, although many steps are routine enough for a much smaller model. Existing ways of sharing this work keep the strong model as the…

    12 upvotes

  • Learning Self-Orchestration in Decentralized Agent Swarmsnew

    The capabilities of an agent swarm depend on both its individual members and how they coordinate. Yet existing multi-agent systems often train agents within fixed orchestration structures, separating the learning of individual capabilities from the organization of collective…

    10 upvotes

  • CarryOn: A Dual-Loop Harness for Personalized and Reliable Robot Assistancenew

    While generalist vision-language-action (VLA) models have enabled diverse open-world manipulation, real-world robotic assistance requires two critical capabilities they currently lack: personalization and reliable execution. First, executing an instruction like "make me a…

    10 upvotes

  • CoDA: Co-Evolution of Data Engines and Action Policiesnew

    Generative simulation can automate robot environment construction, but physically valid environments are not necessarily useful for training. The challenge is to identify environments that address a policy's failures and supervision that helps it overcome them. We introduce…

    9 upvotes

Lobsters

25
Show 17 more

Dev.to top

20
Show 12 more

Techmeme

15
Show 7 more

About this issue

390 items from 77 sources, 6 of them seen in several sources. 7 sources failed to fetch.

Source health

77Which sources were fetched today, and how many items each returned.
SourceSectionItemsStatus
GitHub TrendingMain13Fetched
Product HuntMain50Fetched
Show HNMain4Fetched
HN front pageMain30Fetched
YC LaunchesMain20Fetched
HF trending modelsMain20Fetched
HF trending spacesMain20Fetched
HF daily papersMain20Fetched
LobstersMain25Fetched
Dev.to topMain20Fetched
TechmemeMain15Fetched
AndroidX releasesAndroid · Official & releases47Fetched
Android Developers BlogAndroid · Official & releases3Fetched
Android Studio releasesAndroid · Official & releases2Fetched
Kotlin blogAndroid · Official & releases5Fetched
Kotlin releasesAndroid · Official & releases0Fetched
Compose Multiplatform releasesAndroid · Official & releases4Fetched
Android Developers YouTubeAndroid · Official & releases0Failed
Kotlin on YouTube (JetBrains)Android · Official & releases0Failed
Google Developers BlogAndroid · Official & releases1Fetched
AndroidX merged API changesAndroid · AOSP / AndroidX insider35Fetched
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 · Experts2Fetched
TechYourChance (Vasiliy Zukanov)Android · Experts1Fetched
Kt. Academy (Marcin Moskała)Android · Experts2Fetched
Touchlab (Kotlin Multiplatform)Android · Experts3Fetched
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 releases0Failed
Metro releasesAndroid · Library releases0Failed
Molecule releasesAndroid · Library releases0Fetched
Turbine releasesAndroid · Library releases0Fetched
Paparazzi releasesAndroid · Library releases0Fetched
SQLDelight releasesAndroid · Library releases1Fetched
OkHttp releasesAndroid · Library releases0Fetched
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 releases0Fetched
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 releases0Fetched
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 & ecosystem0Failed
Android AuthorityAndroid · Leaks & ecosystem6Fetched
Android FaithfulAndroid · Leaks & ecosystem1Fetched
Android WeeklyAndroid · Community15Fetched
GitHub Trending · KotlinAndroid · Community16Fetched
Lobsters · androidAndroid · Community3Fetched
HN · AndroidAndroid · Community9Fetched
ProAndroidDevAndroid · Community0Failed
Kotlin WeeklyAndroid · Community1Fetched
Philipp Lackner (YouTube)Android · Community0Failed
1 of 5

Google opens its developer docs to AI coding assistants

Your coding agent can now check current Android docs before it writes code, which should cut outdated API suggestions.

Google launched the Developer Knowledge API and an MCP server in public preview. MCP is the common way coding agents call outside tools. Together they let assistants search and fetch current official documentation for Android, Firebase, Google Cloud and other Google products, in a format machines can read, instead of guessing from what they learned during training.

Google Developers Blog · OfficialRead the source

1 of 4

Anthropic publishes an official directory of Claude Code plugins

Curated plugin lists will decide which agent add-ons most developers ever see.

Anthropic released a GitHub repository listing Claude Code plugins it manages and checks for quality, and it was the fastest-rising project on GitHub's trending page. Cursor published its own plugin specification and official plugins the same day. Both companies are moving from scattered community add-ons to curated lists they control.

GitHub Trending · Fastest risingRead the source