Tools Bench.

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

Last Brew Time: Aug 12, 2026, 10:56 AM PT

Insight

Builders here split into two camps: a long tail repackaging the same open weights over and over for Hugging Face likes, and a sharper minority making real domain bets

01.FINANCE-SPECIFIC FOUNDATION MODELS

02.EDGE-DEVICE FOUNDATION MODELS

03.CODING-AGENT FLEETS

04.ENTERPRISE AGENT PLATFORMS GOING PUBLIC

Featured

GitHub87.5K

infiniflow/ragflow

RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs

youtube
RAGFlow: Free Open Source Generative AI Platform

Market Signal

Why It Has Market Pull

RAGFlow is a mature, widely-adopted open-source RAG-plus-agent engine built by InfiniFlow, a Shanghai-based company backed by Sinovation Ventures and other Chinese venture funds. With nearly 87,500 real GitHub stars, over 10,000 forks, and a very active multi-year issue tracker spanning deployment questions, GPU support, and a public roadmap process, it is one of the most established and battle-tested products in this batch.

  • 87,486 real GitHub stars and 10,307 forks, repo created Dec 12, 2023 and still receiving commits the same day as this review
  • Backed by InfiniFlow (founded 2023, Shanghai), funded by Sinovation Ventures, Capital X, CEC Fund, and Chenhui Venture Partners
  • 1,880 open issues with deep, sustained engagement, one 'ROADMAP 2025' issue alone drew 64 comments of feature requests
  • A single support thread ('[Question]: the process') drew 156 comments, evidence of a large, active troubleshooting community
  • Companion product Infinity (AI-native database for RAG workloads) extends the same company product line

feedbacks

What People Are Saying

  • "Fail to access model(mistral). ERROR: [Errno 111] Connection refused... Ollama is really popular now for local machine."GitHub issue

  • "This is a recurring issue that has been reported across multiple RAGFlow versions (v0.20.3 through v0.24.0). The root cause is a Bearer token prefix handling bug"GitHub issue

  • "Does RAGFlow not have an administrator entrance?"GitHub issue (roadmap thread)

  • "How about supporting multi-modal RAG?"GitHub issue (roadmap thread)

  • "the PyTorch version packaged inside does not support RTX 5090 (sm 90)"GitHub issue

  • "RAGFlow turns the hardest parts of RAG into platform capabilities: complex document parsing, explainable chunking, citation grounding, multi-path retrieval, reranking"Dev blog (knightli.com)

GitHub43.7K

stablyai/orca

Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and VPS.

github
Orca — The AI Orchestrator for 100x builders

Market Signal

Why It Has Market Pull

Orca (Stably AI, YC W22) is a viral, YC-backed open-source Agent Development Environment for orchestrating fleets of parallel coding agents (Claude Code, Codex, OpenCode, and 20+ others) via isolated git worktrees. It has grown explosively to over 43,000 GitHub stars in about five months with a large, global, highly engaged user base filing real production bugs, making it one of the strongest agentic-tooling signals in this batch.

  • 43,717 real GitHub stars and 3,049 forks about 5 months after the repo's March 17, 2026 creation; crossed 31,000 stars within its first four months
  • Y Combinator-backed (Stably AI), MIT licensed, no proxy fees, users bring their own agent subscriptions
  • 3,677 open issues and 30 contributors, reflecting a large, highly active user base filing real bugs (Windows IME crashes, PTY orphaning, remote Linux agent support)
  • Daily-ship release cadence per independent review, with most reported bugs fixed within 24 hours
  • Supports 30+ CLI coding agents (Claude Code, Codex, Gemini, OpenCode, Grok, and more) across desktop, mobile, and VPS

feedbacks

What People Are Saying

  • "If you've been running three terminal panes (Claude Code, Codex, Cursor) and cd-ing between worktrees, Orca is the tool you've been building badly in your head."Dev.to review

  • "Running Claude Code three times in parallel uses 3x your Anthropic quota, new users get caught by surprise."Dev.to review

  • "Even when worktrees solve the filesystem isolation problem, agents can still make semantically conflicting decisions that do not produce merge conflicts."Dev.to review

  • "Korean double consonants (쌍자음) trigger automatic line breaks in the Windows integrated terminal"GitHub issue

  • "Renderer crash+restart orphans open terminal PTYs, crashing shells"GitHub issue

  • "How can I configure and use agents running on the remote server? Is it only possible to use agents on my local computer"GitHub issue

  • "v1.4.35 quits ~400ms after launch on macOS 26.3.1 (Tahoe), single-instance lock acquisition fails on first run"GitHub issue

GitHub36.9K

shiyu-coder/Kronos

Kronos: A Foundation Model for the Language of Financial Markets

youtube
Kronos AI: Tiny Model That Beats Expensive Trading Bots

Market Signal

Why It Has Market Pull

Kronos is a genuinely influential open-source research project -- the first foundation model trained specifically on financial candlestick data -- peer-accepted at AAAI 2026, with fast organic GitHub growth and real quant-trading practitioners fine-tuning and backtesting it. It's a research artifact rather than a company, so there's no funding or commercial signal, but the technical credibility and community pull are unusually strong.

  • 36,900+ GitHub stars and 6,100+ forks, with +6,500 stars gained in a single week per independent trend trackers.
  • Reached #1 on GitHub Trending (July 22, 2026).
  • Accepted at AAAI 2026, with an OpenReview listing and a public Hugging Face demo.
  • Active fine-tuning community: users post real backtest charts and multi-GPU training setups in the issue tracker.
  • 259 open issues on a repo pushed within the last week -- still being actively maintained, not abandoned.

feedbacks

What People Are Saying

  • "实际的预测效果，方向都对了，而且相差也不多。可以拿来写策略啦。"GitHub issue

  • "肯定不是，我自己用ETH的数据，训练的。"GitHub issue

  • "AI 回我：算法不适合预测，适合回顾，用这个预测不如抛硬币"GitHub issue title

  • "the fastest growing GitHub repos in finance this week: 1. shiyu-coder/Kronos (+6.5K stars)"X post

  • "first open-source foundation model for financial candlesticks... predicts OHLCV candles as tokens -- literally GPT for price charts."X post

  • "By open-sourcing a foundation model trained on 45+ exchanges, it rivals systems costing millions to build in-house."Tech blog coverage

GitHub8.7K

Lightricks/LTX-2

Official Python inference and LoRA trainer package for the LTX-2 audio–video generative model.

github
LTX-2 README demo video

Market Signal

Why It Has Market Pull

LTX-2 is a legitimate, actively developed open-source release from Lightricks, an established consumer AI company known for Facetune and Videoleap. The model generates synchronized video and audio in a single pass and has already been ranked among the top image-to-video models by independent benchmarks, backed by rapid, real GitHub growth and ongoing engagement from the company's own engineers in the issue tracker.

  • 8,666 real GitHub stars and 1,394 forks, reached in about 7 months since its January 2026 creation
  • Actively maintained, commits pushed the same day as this evaluation
  • Ranked in the top 3 image-to-video models by Artificial Analysis, behind only Kling 3.5 and Veo 3.1
  • Generates up to 20 seconds of 4K video at 50fps with synchronized audio in a single pass, described in press as an industry first
  • Released under Apache 2.0 with free commercial use for companies under $10M revenue, encouraging real-world adoption

feedbacks

What People Are Saying

  • "I also encountered this. Using the same image but adding "A cinematic scene of ___" changed it from a static ... slide to an actual video. It's quite strange."GitHub issue

  • "I jinxed myself now half my videos are just static."GitHub issue

  • "The model needs to be reloaded for each prompt during testing, which slows down the process."GitHub issue

  • "Thank you for your feedback we will consider this in the future updates."GitHub issue (Lightricks team reply)

  • "Okay I finally got it to run with the original full model but the 25 sec took 12 hours in the second stage."GitHub issue

  • "Same question. Thanks for LTX Team!"GitHub issue

GitHub4.1K

cactus-compute/needle

14MB foundation model for tiny devices; phones, wearables, smart home, and robots.

youtube
Cactus Needle - The 26M Function Calling Model

Market Signal

Why It Has Market Pull

Needle is a real, funded product from Cactus Compute, a Y Combinator (Summer 2025) startup that raised seed funding backed by Oxford Seed Fund and Google for Startups to build efficient on-device AI. The 14MB agentic model drove a very high-engagement Hacker News launch with deep technical back-and-forth involving the founding team, plus genuine third-party integration work in the Home Assistant community, and the GitHub repository shows fast, real growth since its February 2026 release.

  • 4,109 real GitHub stars and 301 forks, reached in about 6 months since February 2026 creation
  • Actively maintained, commits pushed the same day as this evaluation
  • Cactus Compute is YC-backed (Summer 2025 batch) with seed funding from Oxford Seed Fund and Google for Startups
  • Companion Show HN launch for Needle 2 scored 525 points with dozens of detailed technical comments, including live responses from Cactus team members
  • Independent integration thread in the Home Assistant community forum exploring Needle 2 for on-device voice assistant use

feedbacks

What People Are Saying

  • "Roman from Cactus here - yes you're right, there's only so much a 14MB model can do. Needle excels at in-context inference, with tightly defined environments."HN comment (team)

  • "it did a pretty good job for tool calling when I tried it, and I think it could be pretty useful"HN comment

  • "This is extremely impressive if it works."HN comment

  • "Seems like a match made in heaven for hass on-device voice assistant usage"Home Assistant community forum

  • "I was already thinking of compressing functiongemma-270m-it down to 1-2 bits so it would work flawlessly in the browser. Your Fine-tuning feature is even much more convenient."HN comment

  • "Funny result from the web demo. I'm well aware that it's an extremely small and, well, stupid, model, but even so..."HN comment

Sources

GitHub

msitarzewski/agency-agents

144.4K

A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.

NanmiCoder/MediaCrawler

61.9K

小红书笔记 | 评论爬虫、抖音视频 | 评论爬虫、快手视频 | 评论爬虫、B 站视频 ｜ 评论爬虫、微博帖子 ｜ 评论爬虫、百度贴吧帖子 ｜ 百度贴吧评论回复爬虫 | 知乎问答文章｜评论爬虫

hugohe3/ppt-master

45.4K

AI turns documents or topics into real, native PowerPoint decks—with native shapes, transitions and animations, data-backed charts and tables on demand, audio narration from speaker notes, and support for your own .pptx templates. · by Hugo He

Product Hunt

Tines 3B

412

Tines 3B is the single, secure environment for your most important agents, apps, and automation. Build with AI, from anywhere. Code runs isolated and credentials stay protected. Everything you ship is fully auditable and monitored from one place. Tines 3B is code-first and AI-native, built to help you move fast without compromising on security. Explore Edition is available today and gives you access to 3 live workflows with unlimited users, spaces, and connectors.

BetterClaw

321

BetterClaw is a no-code platform for building AI agents that run on a schedule without you. Connect Gmail, Slack, or Telegram, and your agent triages inbox, sends morning briefings, or monitors what you care about. Agents start as Interns that ask before acting. Bring your own AI key and it's genuinely $0.

Xirp

276

The agentic development environment with institutional memory. Xirp connects to your services, ownership, docs, and architectural decisions so every session starts with real context. Powered by Spotify Portal.

Equitybee Benchmark

247

Equitybee Benchmark helps U.S. startup employees understand how their equity grants compare. Explore 9,000+ verified new-hire grants across 2,500+ startups by department, seniority, and company stage. For years, companies have benchmarked the equity they offer. Now you can benchmark yours. -- The benchmark provides market context based on Equitybee's dataset and is intended for informational purposes only. The dataset has not been independently verified. Data presented as is.

Bullet

227

Bullet is a coding agent built for speed. We got tired of burning hours waiting on agent runs. The models were fine, but the loops around them were slow. Bullet auto-picks the right model/reasoning level per prompt, parallelizes searches/reads/commands, and uses targeted code search instead of embedding your whole repo. Works with your Claude Code or Codex subscription, API keys, or an on-device model. 95.8% on SWE-bench Verified (top 3), 119s/task. Built by Yale CS grads, ex-AppLovin/Citadel.

bb

189

bb is an agentic orchestrator GUI, not unlike the Codex app, but it works with any provider: Claude Code, Codex, OpenCode, and more. What makes bb different is that it can customize and extend itself. Almost anything in bb can be changed with a single prompt. Ask for a task tracker, and one appears. bb also creates a skill that teaches all of your agents how to use it. Instead of waiting for your agentic workspace to add the feature you need, you can just ask for it.

YC Launch

Hacker News

Needle2: 14MB agentic LLM for phones, wearables, smart home and robots

524 pts

Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits betwee... (524 points, 176 comments).

OJCP – an open protocol for agent-consumable job data

8 pts

Author here! Agents are applying to jobs for people right now, with progressively more volume, and there's nothing built for it. So they scrape career pages and fight ATS forms with Playwright/Browser Use, which breaks constantly (or they get bot blocked). Employers get buried in applications that don't fit, candidates hear nothing back, and the resume is now an AI-written thing that another AI scores (which breaks the existing model entirely, btw). OJCP is MCP tools for search and apply, a mani... (8 points, 0 comments).

Lumabri – What if LLMs worked like Napster?

8 pts

A while ago I started working on Colibrì to see if it was possible to run huge LLMs on a normal computer. The project grew far beyond what I expected, thanks in large part to the HackerNews community. That led me to a new question: What if we stopped thinking about one computer? This is the idea behind Lumabri. Instead of requiring a single machine to store and run an entire huge model, Lumabri treats a network of normal computers as a shared pool of resources. One machine might provide disk spa... (8 points, 10 comments).

HeyDeacon, AI customer support agent for SaaS founders

4 pts

If you're launching a new webapp, pretty quickly you'll need a way to manage customer queries, onboarding, collate user feedback and track actions with analytics. Deacon is a drop in SDK that does that for you. You can now speak with every site visitor, collate qualitative feedback and use it to iterate and grow faster. Let me know what you think (4 points, 1 comments).

HF Spaces

Qwen-Image-Edit-2511-LoRAs-Fast

2.5K likes

Demo of the Collection of Qwen Image Edit LoRAs Qwen-Image-Edit-2511-LoRAs-Fast is a Hugging Face Space tagged with gradio, mcp-server, region:us. It has 2475 likes on Hugging Face.

MiniMax H3

235 likes

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...

CharacterSheet LoRA Demo

125 likes

Multi-view character sheet from one image (FLUX.2 LoRA) This Space demonstrates the CharacterSheet QuadView LoRA applied on top of FLUX.2 Klein 9B. Upload a clear, well-framed image of a character and the model produces a multi-view character sheet: a face close-up plus front, side, and back full-body views on a single 1536×1024 sheet. CharacterSheet LoRA Demo is a Hugging Face Space tagged with gradio, mcp-server, region:us. It has 125 likes on Hugging Face.

MiniMax-H3 Ultra Fast

80 likes

Ultra-fast local NVFP4 video + synchronized audio generation MiniMax-H3 Ultra Fast — local conditioner + pruned NVFP4 on Blackwell Joint video and synchronized sound from MiniMax-H3, rebuilt for a single 96 GB Blackwell ZeroGPU worker. | layer | optimization | |---|---| | Weights | 12.5 GB pruned NVFP4 transformer: 20.1B effective parameters instead of 33.1B/61.7 GiB BF16. | | Compute | Native CUDA 13 NVFP4 tensor-core GEMMs through comfy-kitchen; higher-precision norms, embeddings and output heads. | | Residency | Transformer, conditioner and both VAEs remain GPU-resident during generation—no layerwise CPU offload. | | Conditioner | Local 15.7 GB Qwen3-VL NVFP4-AWQ checkpoint containing onl...

MiniMax H3 Turbo LoRA

58 likes

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...

Wan2.2 14B Fast Preview

1.1K 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 1121 likes on Hugging Face.