Daily Digest – 2026-07-22

psf/black

Stop debating Python formatting; this just decides it for you and moves on. (Python, 41.7k★)

★ 41,800

tokio-rs/tokio

Rust async runtime that doesn't make you want to flip your desk over. (Rust, 32.6k★)

★ 32,821

NVIDIA/Megatron-LM

Train transformer models at scale without selling a kidney for compute. (Python, 17.2k★ ⚡25★/day)

★ 17,165

zitadel/zitadel

Self-hosted identity that works instead of making you rewrite auth from scratch. (Go, 14.5k★)

★ 14,510

EleutherAI/lm-evaluation-harness

Benchmark your LLM against standard evals without building your own test suite. (Python, 13.4k★)

★ 13,371

bytedance/seedream-5-pro

ByteDance's text-to-image that hits 1K/2K resolution without looking like a fever dream. (27K runs)

bytedance/seedance-2.0-mini

Lighter video model for high-volume generation; cheaper tokens, acceptable quality. (26K runs)

csviverdeia/locateanything-3b-h100

LocateAnything-3B em H100 (Hopper) — visual grounding imagem+video (boxes/points). Detecção densa, pointing, tracker none/sort/reid. PRINCIPAL (rápido). (23K runs)

ultralytics/yolov8s-worldv2

Open-vocabulary object detection; tell it what to find instead of training on fixed classes. (17K runs)

prunaai/p-image-try-on

Virtual try-on that doesn't look like a glitch; clothes on bodies, actually convincing. (14K runs)

reader-lm

HTML to Markdown model; clean your web scrapes before feeding them to downstream tasks. (898.9K pulls)

shieldgemma

Safety evaluator for prompts and outputs; check your AI isn't about to say something stupid. (897.4K pulls)

deepcoder

14B coding model at O3-mini level; local code gen that doesn't need Anthropic's API. (887.9K pulls)

llama-pro

Llama 2 expanded for code and math; trade generality for domain chops. (881.8K pulls)

yarn-mistral

Mistral with 64K/128K context; long documents fit in your inference window now. (875.9K pulls)

@opentelemetry/instrumentation-openai

Trace your OpenAI calls without reinventing the wheel; hooks right into OTel. (5.2M/wk)

@tanstack/ai-client

Framework-agnostic chat client with streaming; use it everywhere, own the integration. (121K/wk)

genkit-cli

CLI for Google Genkit; orchestrate AI workflows without learning another vendor's DSL. (93K/wk)

@genkit-ai/telemetry-server

Monitor Genkit AI workflows locally; debugging that doesn't phone home. (89K/wk)

repomix

Pack your repo into a single file for LLM context; feed your codebase to Claude painlessly. (80K/wk)

vllm

Serve LLMs faster and cheaper than you thought possible; throughput that makes sense. (1.3M recent downloads)

unstructured

Convert PDFs, HTML, images into text without losing your mind or data integrity. (1.2M recent downloads)

together

Official Together API client; access open models like you're not being price-gouged. (576K recent downloads)

outlines

Constrain LLM outputs to follow rules; JSON, regex, grammars that actually work. (491K recent downloads)

agno

Programming language built for agents; if you're building agent swarms, try this first. (471K recent downloads)

jbilcke-hf/FacePoke_CLONE-THIS-REPO-TO-USE-IT

Face-swapping demo that runs in browser; clone it, use it, don't ask questions. (docker, 2213 likes)

selfit-camera/Omni-Image-Editor

Web-based image editor with layers and filters; Photoshop you don't have to pay for. (gradio, 2189 likes)

tencent/Hunyuan3D-2.1

Text-to-3D model that doesn't require a PhD in prompting to get usable outputs. (gradio, 2153 likes)

finegrain/finegrain-image-enhancer

Upscales and sharpens images without the AI jank you usually get. (gradio, 2151 likes)

Wan-AI/Wan2.1

Video generation model; it's fast enough to not make you reconsider your career. (gradio, 2138 likes)

jinaai/jina-embeddings-v3

Embeddings model that understands retrieval better than your last BERT attempt. (feature-extraction, 3.1M downloads)

timm/resnet50.a1_in1k

Production-ready ResNet50 that's been tuned to not embarrass you in production. (image-classification, 3.1M downloads)

jonatasgrosman/wav2vec2-large-xlsr-53-polish

Speech-to-text for Polish that actually works; multilingual ASR without the enterprise markup. (automatic-speech-recognition, 3.0M downloads)

pyannote/voice-activity-detection

Detects when someone is actually talking versus silence; dumb simple, actually works. (automatic-speech-recognition, 3.0M downloads)

Qwen/Qwen3-Embedding-8B

8B embedding model that's smaller than your startup's tech debt but still competitive. (feature-extraction, 2.9M downloads)

Maximilians/ps2_hf1

PlayStation 2 game dataset; niche, but if you need it you need it. (545K downloads)

angie-chen55/python-github-code

1M+ Python repos for training models; free code corpus that took someone forever to collect. (542K downloads)

applied-ai-018/pretraining_v1-omega_books

100M+ books dataset for pretraining; bulk reading material for your LLM. (536K downloads)

Humalike x Hermes

Social intelligence plugin for Hermes Agent (299 upvotes) [producthunt]

Buzzy

Your creative AI co-director (273 upvotes) [producthunt]

Kastra

Runtime authorization for Claude, Cursor, Codex and OpenClaw (224 upvotes) [producthunt]

Jockey by TwelveLabs

The video AI agent that understands your whole library (210 upvotes) [producthunt]

box

Simple computers for agent w/ full VMs (197 upvotes) [producthunt]

Justif – Knuth-Plass justification and microtypography for the web

Show HN: Justif – Knuth-Plass justification and microtypography for the web (186 upvotes) [hackernews_showhn]

Skim

Free, open-source AI email client for Windows (176 upvotes) [producthunt]

Manifest

Turn any webpage into an action manifest for AI agents (156 upvotes) [producthunt]