Tools Bench.

Product launches and open-source repos with enough signal to earn a second look.

Last Brew Time: Sep 12, 2026, 11:48 AM PT

Insight

This run's clearest signal is that shipping fast now means standing on someone else's checkpoint

01.VIDEO-MODEL REHOSTS MULTIPLYING OVERNIGHT

02.BROWSER-NATIVE HARDWARE AND ROBOTICS SIMS GOING MAINSTREAM

03.CODING-AGENT HARNESSES RACING TO OUT-ORCHESTRATE EACH OTHER

Featured

HF Spaces506 likes

Microduck Sandbox

Real trained RL policies for the Microduck robot, running fully in the browser: MuJoCo compiled to WebAssembly steps the physics, onnxruntime-web runs the policy network at 50 Hz. No server, no backend. Two locomotion variants of the same robot are included: legs (walking, the default) and rollers (the wheeled skating variant). Press M (or hold D-pad up ~1 s on a gamepad) to switch; the roller model, meshes and policies are lazy-loaded on the first switch. | Mode | Checkpoint | What it does | |--------|-----------|--------------| | Run (legs) | BESTalphawalking.onnx | Velocity-tracking locomotion (arrows / WASD to steer) | | Sit | BESTalphasitstand.onnx | Sits down on its hull, stands back u...

youtube
Microduck Sim 2 Real.

Market Signal

Why It Has Market Pull

This is a real product from a funded robotics company now owned by Hugging Face, not a hobby demo: the browser sandbox runs the same trained policies shipping on a $399 physical robot going into production this year. Press coverage was immediate and broad, and the underlying code already has a large open-source following.

  • 506 Hugging Face likes plus coverage from Engadget, The Register, and MarkTechPost within the same week of the robot's announcement.
  • Built by Pollen Robotics, a robotics company founded in 2016 that raised roughly $2.8 million before being acquired by Hugging Face in 2025.
  • Companion GitHub repositories already show 8,173 stars (core robot software) and 2,079 stars (reinforcement-learning environments).
  • Runs the real trained neural-network policy fully client-side at 50Hz via WebAssembly physics and onnxruntime-web, no backend required.
  • $399 physical robot is open for pre-order with deliveries targeted before Christmas 2026.

feedbacks

What People Are Saying

  • "Try Microduck routines in the simulator using the robot's API?"HF community comment

  • "Microduck Racer - Simulator modified for racing"HF community comment

  • "Tiny open-source robot duck can walk and learn from its mistakes"tech press coverage

  • "Hugging Face and Pollen Robotics open pre-orders for the $399 Microduck"tech press coverage

  • "Hugging Face offers $399 robot duck to help you quack the AI code"tech press coverage

  • "Hugging Face's Pollen Robotics Releases Small, Cute & Open-Source 'Microduck' Robot"tech press coverage

HF Spaces418 likes

MiniMax H3 Turbo LoRA

Video generation with a synchronized soundtrack MiniMax-H3 — unquantized, split across two Spaces Joint video and soundtrack out of a single denoising pass, at bfloat16 with no quantization anywhere. This Space is the denoising half: the 61.73 GiB transformer and the two autoencoders. The 62.14 GiB Qwen3-VL conditioner runs in qwen3vl-conditioner, which this Space calls over the gradio API for every request. The weights are the public MiniMaxAI/MiniMax-H3 diffusers checkpoint. MiniMax-H3 is 195.9 GiB in bfloat16 and a ZeroGPU Space is evicted at 150 GB of storage. An unquantized single Space is therefore impossible, which is why quantized demos of it run NVFP4 or float8 weights. Cut the Mini...

youtube
MiniMax H3 Turbo + Custom Frames + Video Extend in ComfyUI

Market Signal

Why It Has Market Pull

MiniMax is a major, well-funded Chinese AI lab — it raised roughly $2 billion in 2026 on top of a prior $300M round, listed in Hong Kong, and reports around $400M in annualized revenue. This Space is a first-party demo of its own H3 video-plus-audio model with a speed-optimized LoRA, and the underlying model has genuine large-scale developer adoption, not just marketing buzz.

  • MiniMax raised about $2 billion in 2026 funding and reached roughly $400M in annualized revenue — a real commercial company, not a hobbyist reupload.
  • The parent MiniMax-H3 model has 5,190 likes and 4.86 million downloads in the last month alone on Hugging Face, plus 104 community discussion threads and over 100 derivative Spaces.
  • This specific Turbo-LoRA Space has 418+ likes and 10 active discussion topics, including a merged community pull request — evidence of real back-and-forth engineering engagement, not just passive traffic.
  • The Turbo LoRA cuts required sampling steps roughly 5x (to as few as 4 steps) for joint video-plus-audio generation, a concrete, benchmarked speed win over the base model.
  • Version history shows disclosed, iterative quality fixes (the v4 release resolved the 'over-sharpening / plastic look' of v1) alongside honestly-stated remaining weaknesses (motion smear on fast, large motion at 4 steps).

feedbacks

What People Are Saying

  • "OAuth sign-in missing inference-api and write-repos scopes — canvas workflow can't start"HF community discussion title

  • "fix: correct inference-api scope typo in YAML"HF community pull request

  • "This branch is ready to get merged automatically"HF community pull request

  • "v4 checkpoint demonstrates much better static and small-motion shots, markedly better micro-detail (faces, fingers, fine texture)"HF community discussion

  • "the over-sharpening / plastic look of the earlier v1 line is fully resolved"HF community discussion

  • "at 4 steps with large, fast motion, v4 can produce motion-smear / trailing ghosting"HF community discussion

Product Hunt317

Raycast 2.0

Raycast 2.0 is built on a new foundation, redesigned from the inside out. Your shortcut to everything now brings AI that can take action across your apps, Automations for recurring tasks, and Projects to keep ongoing work together. Connect your own ChatGPT or Claude account and put AI to work alongside the commands and extensions you use every day. The best Launcher got even better.

youtube
102 Things You Can Do With Raycast 2.0

Market Signal

Why It Has Market Pull

Raycast 2.0 is a major relaunch from an already-established, well-funded productivity company with a large existing user base — this is proven momentum extending into agentic AI, not a speculative new entrant.

  • Company has raised roughly $47.8M total, including a $30M round backed by well-known venture investors
  • Hundreds of thousands of daily users and over 20,000 extension developers on the existing platform
  • The 2.0 launch ranked #4 with 318 upvotes, and the product carries a 4.9 out of 5 rating across 160 reviews
  • Publicly described by the team as its biggest release since the company launched in 2020

feedbacks

What People Are Saying

  • "Raycast is the one tool I can't live without! It's so embedded in my day to day that makes my life significantly easier."Product Hunt review

  • "absolute fan of what...and team are building at Raycast. great team, great product."Product Hunt comment

  • "2.0 is an upgrade in the best ways: preserving the core of speed, utility, and flexibility."Product Hunt comment

  • "quickly becomes part of your workflow"Product Hunt review

  • "one of the biggest workflow upgrades for Mac users"Product Hunt review

  • "Product Hunt is about consistency. Take @raycast. 15 launches in 6 years."X reply

Product Hunt256

Cline Desktop App

Cline Desktop gives you an open-source workspace to put your favourite model to work - run multiple agent sessions, automate recurring tasks, extend their capabilities through Cline’s Marketplace, and continue work from agents like Claude Code and Codex. Choose the models and providers that work best for you, while Cline’s agent harness, optimized for open-weight models, powers the work behind them. All without being tied to a closed ecosystem.

Market Signal

Why It Has Market Pull

Cline is a mature, widely used open-source coding agent with real venture backing, and its new desktop app extends that established base rather than launching cold.

  • Core code repository has roughly 68,000 stars and 7,300 forks
  • Raised $32M total, including a $27M round led by a well-known enterprise-software investor at a $110M valuation
  • Desktop app launch ranked #5 with 256 upvotes
  • Long-running community discussion calls it 'hands down, the most effective agentic coding tool out there'

feedbacks

What People Are Saying

  • "oss ftw!"Product Hunt comment

  • "100% . Open source is a big part of what makes Cline, Cline."Product Hunt comment

  • "The open source + model agnostic approach is a strong combination. Developers get more control over both the agent and the models underneath it."Product Hunt comment

  • "This feels less like another coding assistant and more like an environment for managing AI assisted development."Product Hunt comment

  • "The desktop app is really fun to use! I'm enjoying it."Product Hunt comment

  • "Cline is, hands down, the most effective agentic coding tool out there"HN comment

Sources

GitHub

asgeirtj/system_prompts_leaks

65.3K

Extracted system prompts from Anthropic - Claude Fable 5.1, Opus 5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-6-Astra, Codex. Google - Gemini 3.8 Flash, 3.1 Pro, Antigravity. xAI - Grok, Grok Bot, Cursor, Kimi and more! Updated regularly.

bilawalsidhu/gods-eye-view

29.3K

A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe.

vxcontrol/pentagi

23.3K

Fully autonomous AI Agents system capable of performing complex penetration testing tasks

max-sixty/worktrunk

7.2K

Worktrunk is a CLI for Git worktree management, designed for parallel AI agent workflows

Product Hunt

Anysite.io

457

Anysite.io is a B2B data layer living in your favorite AI agent. Ask your agent for the list you need — companies by geo, industry, and size; the people inside them; their current job titles and emails — and Anysite.io returns it. No queries to write, no scrapers to maintain. Works over MCP or REST API inside Claude, Codex, Cursor, OpenCode, and other agents. Plug in and launch your outbound data collection the same day.

Devin Voice

120

Devin Voice lets you talk to Devin, Cognition's AI software engineer, and have it ship the work. Speak a task out loud and Devin plans, codes, and delivers. Powered by GPT-Live for natural conversation, with Cognition's new SWE-2 coding model under the hood. Try Devin at https://devin.ai.

sizeless

93

sizeless turns a smartphone video of an open trench into the documentation utilities and contractors are legally required to produce: a 3D model, CAD/BIM plans, and the quantities they bill from. Today that takes months and a surveyor. We do it in hours, from a video the crew films themselves.

Loqua

402

Your thoughts shouldn’t have to slow down for a keyboard. With Loqua, you can speak naturally to turn rough ideas into ready-to-use writing, understand what’s on your screen, listen when you’d rather not read, and move work forward by voice - from rewriting and translation to scheduling and coding workflows. Less typing, less context switching, more time in flow.

Wisry

313

AI agents that run the whole play: scan the Meta and TikTok ad libraries for the ads already winning in your market, clone them into your brand as static and video ads, and launch to Meta and Google, optimized for ROAS. Winning ads, without the guesswork.

easyspecs.ai

153

The gap between 1.5x and 100x isn't speed, it's trust!! AI agents generate code 100x faster than humans write it. But your team can't review it all. EasySpecs documents undocumented codebases and turns them into trustworthy specifications, so spec review becomes your new merge request review. Ground your agents in reality. Scale agentic development without babysitting. Document the foundation once. Review specs including Oracles and Rubrics and Ship code you actually verified.

YC Launch

Isengard Industries Inc

5

AI Weapons of Mass Production Isengard Industries Inc · Summer 2026 · Industrials Tags: Artificial Intelligence, Swarm Robotics, Unmanned Vehicle. Website: http://isengardindustries.com/

Hacker News

Open-source simulation testing infra for voice agents

15 pts

Hey HN, we’re Nischal & Naman. We’re brothers, and together we’re building an open-source platform for simulation based testing of voice agents (try it out in 5 mins - https://docs.egma.ai/docs/get-started/quickstart , 2 min demo video - https://youtu.be/wgDWEe5UAUY ) Platforms that help you do simulation testing already exist. But they all charge a heavy premium on top of inference costs. We believe if the industry truly wants to scale simulation testing of voice agents, we need to stop chargin... (15 points, 6 comments).

HolaOS––An Opensourced workspace that alternative to Claude

4 pts

Open-source agentic workspace enterprises can make their own. Connect the systems you already run — 100+ integrations, MCP, chat tools, apps, browser, local files — with shared memory. Any agent (Claude Code, Codex), any model, or BYOK. Set up in clicks, not months. Local-first: your data never leaves your machines. (4 points, 0 comments).

Godot and Rust based multiplexer (terminal panes and more)

91 pts

I wanted to share my side-project: gPTY. Started off as an idea to combine Godot and Rust in a project (two stacks I wanted to use more to learn more). The base inspiration was tmux - simply allow spawning multiple PTYs and then let the user grid/tile them how they see fit. But since we have the Godot game engine at our disposal, we can do some more interesting things, like add an FPS counter, and then subsequently also let people set their preferred FPS (the idea being the potential lower power... (91 points, 47 comments).

Extension to filter LLM written articles

16 pts

Firefox Extension/Userscript and API to get Pangram scores for all articles on the hackernews frontpage. The extension allows you to hide articles with a high score. This is about detecting posts written by LLMs, not posts about AI. Feel free to use the API to build your own tooling/readers. Big thanks to https://news.ycombinator.com/user?id=salahadawi for providing the data :). (16 points, 2 comments).

ShevtoneAudio Orchestrator – Turning MIDI into Full Orchestration

6 pts

Hi HN, I'm Bor Shev, a composer and developer. Over the past two years I've been developing ShevtoneAudio Orchestrator. The idea is simple: instead of generating a finished piece of music and replacing the composer, Orchestrator takes the composer's own MIDI and develops it into a full orchestration. It analyzes the musical material — harmony, melody, rhythm, dynamics, structure and orchestral density — and creates an arrangement across strings, brass, percussion and other sections. The importan... (6 points, 4 comments).

Dbmask – Discover, mask, and verify sensitive data in SQL databases

5 pts

Hi HN I developed *dbmask*, an open-source Python tool designed to discover sensitive columns in SQL databases, masking them with deterministic fake values, and validate the masking results. The motivation behind this tool stems from a common issue I encountered when handling test data: existing masking tools often assume the user already knows exactly which columns contain sensitive information. Furthermore, the successful completion of a masking task does not guarantee that all sensitive value... (5 points, 0 comments).

HF Spaces

H3 Acceleration Arena

99 likes

Blind A/B ranking of MiniMax-H3 acceleration variants Human-judged ranking of ~26 MiniMax-H3 acceleration variants over a 200-prompt corpus, from blind pairwise votes on pre-generated clips, with confidence intervals, cost and slice breakdowns. The design and its reasoning are in arena/DESIGN.md; the app's own notes are in arena/README.md. This Space is private and must stay private until deliberately flipped. It streams ~3,700 clips out of the private dataset multimodalart/h3-pre-gen-arena. See Going public below. hfoauth: true above creates the OAuth app and injects OAUTHCLIENTID, OAUTHCLIENTSECRET, OAUTHSCOPES and OPENIDPROVIDERURL. arena/space_auth.py implements the flow by hand (this is...

Fruit Fly Simulation

78 likes

Simulate a fruit fly in your browser using WebGPU kernels A standalone fruit fly connectome demo: paint neurons, stimulate the network, and watch an articulated Three.js fly respond. Vanilla JavaScript, Vite, Three.js, and @huggingface/kernels. Simulated neural activity drives crafted walking, turning, and flight animations. The movements are illustrative, not validated predictions of fly behavior. Open the address printed by Vite, then click Download & start. All neural weights, fly meshes, fonts, and kernel templates are included in public/; setup loads them into the browser and caches verified weight chunks. No API key or external model service is required. Deploy the generated dist/ dire...

Viggle-Animate

51 likes

Put the person from a still into a driving video, in 4 steps Upload a driving video and a character still. The clip's first frame is repainted with gpt-image-2 so it shows your character in that exact pose and framing, and that repainted frame is the only conditioning the model gets besides the clip itself — no pose estimation, no segmentation, no masks. Takes 4–10 s of driving video and renders it at 4 sampling steps; anything longer is cut to 10.1 s. The model itself only renders frame counts of the form 17k+5 and never fewer than 124 (5.17 s), so shorter clips are held out to 124 frames with a frozen last frame and the render is trimmed back — what comes out is exactly as long as what wen...

Wan2.2 14B Fast Preview

2.0K likes

generate a video from an image with a text prompt Wan2.2 14B Fast Preview is a Hugging Face Space tagged with gradio, mcp-server, region:us. It has 2035 likes on Hugging Face.

FLUX.2 Klein multi-LoRA

553 likes

Use multiple FLUX.2-Klein LoRAs Note: This space is experimental and may log image-uploads during certain periods for performance monitoring. Always comply with HF and model Terms of Service. Any stored images are automatically deleted after 7 days. FLUX.2 Klein multi-LoRA is a Hugging Face Space tagged with gradio, mcp-server, region:us. It has 553 likes on Hugging Face.

Qwen-Image-Edit-2511-LoRAs-Fast

270 likes

Demo of the Collection of Qwen Image Edit LoRAs Prompt-based image editing on Qwen-Image-Edit-2511, with identity preservation (change pose/clothes/scene while keeping the person's face), HD output, and an anti-deformed-hands option. The Space exposes one endpoint: /edit. Base URL / Space id: kulkas2pintu/QWENEDITIMAGE. | name | type | default | meaning | |---|---|---|---| | image | image (filepath/URL) | — | required. The image to edit (the person/subject whose identity is kept). | | prompt | str | — | required. What to change, e.g. "change her outfit to a red dress; turn to look left". | | loraadapter | str | "" | Optional style LoRA name (see list below). "" = no style / base edit. | | im...

AI Tools — September 12, 2026 Edition | Agentic Brew