Daily Digest – 2026-07-17

pingcap/tidb

Distributed SQL database with ACID, vector search, and no infrastructure ceiling for agentic workloads. (Go, 40.3k★)

★ 40,301

SigNoz/signoz

OpenTelemetry observability stack with logs, metrics, traces, and native AI agent debugging in one place. (TypeScript, 30k★)

★ 29,975

deepspeedai/DeepSpeed

Distributed training and inference optimization library; makes large model training actually feasible on limited hardware. (Python, 42.7k★)

★ 42,729

OpenAPITools/openapi-generator

Auto-generate API clients, server stubs, and docs from OpenAPI specs; saves weeks of boilerplate per integration. (Java, 26.5k★)

★ 26,535

dapr/dapr

Portable runtime for distributed apps; event-driven architecture plus workflow orchestration without vendor lock-in. (Go, 26k★ ⚡9★/day)

★ 25,957

prunaai/p-image-edit

Sub-1-second multi-image editing model at $0.01 per call; actually production-ready for high-volume features. (36.3M runs)

openai/gpt-image-2

State-of-the-art image generation from text; sharp text rendering and instruction-following beats competitors. (14.5M runs)

google/gemini-3-flash

Fast frontier model with search grounding; Gemini's speed tier for latency-sensitive apps without sacrificing quality. (5.5M runs)

prunaai/p-video

Fast video generation with draft mode; text/image/audio-to-video in one endpoint, production cost-effective. (2.2M runs)

google/gemini-3.1-pro

Google's flagship reasoning model with extended thinking; better for complex problem-solving than speed-tier competitors. (1.4M runs)

devstral

24B coding agent model that runs locally; best open-weight option for code generation without API costs. (981.7K pulls)

wizard-math

Open model family tuned for math and logic; 7B to 70B options for different inference budget constraints. (976.6K pulls)

dolphincoder

Uncensored coding model based on StarCoder2; 7B and 15B variants for on-device code generation. (973.5K pulls)

llama3-gradient

LLama3 extended to 1M context tokens; handle entire codebases or documents in one pass without chunking. (966.3K pulls)

aya-expanse

Cohere multilingual models across 23 languages; 8B and 32B options for global user bases. (965.8K pulls)

pac-proxy-agent

PAC file proxy support for Node.js HTTP requests; handle corporate proxies without hardcoding credentials. (25.5M/wk)

puppeteer-core

Headless browser automation without bundled Chromium; use your own browser binary, save install size. (14.5M/wk)

global-agent

Set HTTP/HTTPS proxy globally via env vars; one line replaces proxy config scattered across dependencies. (8.7M/wk)

puppeteer

High-level headless Chrome automation bundled with browser; easy setup for scraping, testing, screenshots. (8.3M/wk)

cypress

Modern frontend testing tool with real browser, fast feedback, zero flakiness; better DX than Selenium. (6.5M/wk)

langchain

LLM composition framework; chains, memory, agents in one package but increasingly bloated for simple use cases. (76.7M recent downloads)

weaviate-client

Python client for Weaviate vector DB; semantic search and RAG without managing infrastructure. (53.0M recent downloads)

datasets

HuggingFace datasets library; load, process, and version datasets at scale for ML without rolling your own. (31.0M recent downloads)

mistralai

Official Mistral API client; drop-in replacement for OpenAI client with cheaper inference. (10.6M recent downloads)

accelerate

HuggingFace library for distributed training; multi-GPU, multi-node without writing boilerplate. (6.6M recent downloads)

zerogpu-aoti/wan2-2-fp8da-aoti-faster

wan2-2-fp8da-aoti-faster (gradio, 3325 likes)

akhaliq/anycoder

anycoder (docker, 3298 likes)

HuggingFaceTB/smol-training-playbook

smol-training-playbook (docker, 3240 likes)

mukaist/Midjourney

Midjourney (gradio, 2986 likes)

pharmapsychotic/CLIP-Interrogator

CLIP-Interrogator (gradio, 2974 likes)

openai/whisper-large-v3-turbo

Fast speech-to-text in 99 languages; smaller than v3 but still crushes accuracy for builders on a budget. (automatic-speech-recognition, 8.0M downloads)

openai/gpt-oss-20b

20B open-weight GPT model; runs locally and on cheaper inference, actual alternative to proprietary APIs. (text-generation, 7.2M downloads)

Qwen/Qwen2.5-3B-Instruct

Tiny 3B instruction-tuned model that actually works; fits in memory and runs fast enough for real-time features. (text-generation, 7.0M downloads)

dima806/fairface_age_image_detection

Vision transformer fine-tuned for age estimation from images; ready-to-use without training your own model. (image-classification, 6.4M downloads)

openai/whisper-base

Lightweight Whisper model for speech-to-text when you need speed over accuracy; trade-off worth it for some use cases. (automatic-speech-recognition, 6.3M downloads)

Salesforce/wikitext

Large text corpus for language model pretraining or fine-tuning; real data at scale without licensing headaches. (1.4M downloads)

banned-historical-archives/banned-historical-archives

Historical images dataset; niche but useful for researchers working with archived visual materials. (1.3M downloads)

xlangai/ubuntu_osworld_file_cache

Ubuntu system state snapshots for agent training; helps agents learn to interact with real OS environments. (1.2M downloads)

V2Fun

Generate 3D character with 8K textures and AI motion capture (670 upvotes) [producthunt]

Velo 3.0

AI video infrastructure to explain, train, and sell faster. (640 upvotes) [producthunt]

Campus

One project space for humans and AI agents (443 upvotes) [producthunt]

Agently

Your whole stack, running itself! (340 upvotes) [producthunt]

Zro

Private inference for coding agents (328 upvotes) [producthunt]

Crustdata Recruiter

Claude Skills to turn Claude into a 100x Recruiter (286 upvotes) [producthunt]

Albato AI

Build AI-driven workflows across 1,000+ apps (207 upvotes) [producthunt]

YAGNI

Proactive agent teams you manage like humans (196 upvotes) [producthunt]