Independent Column

Open-source projects worth exploring

Runnable, documented projects worth studying and extending.

AI and AgentsWarp - AI-native Agentic development environmentPW
Terminals, agents, development environments, Rust, coding, Warp, OpenAI

Warp - AI-native Agentic development environment

Warp is an agentic development environment evolved from endpoints, with built-in encoding agents and support integration with third-party tools such as Claude Code, Codex, and Gemini CLI. Sponsored by OpenAI and open source, built in Rust, 11,955 Stars were added today.

7290
View repository
OthersMiroFishPW
Collective intelligence, prediction, multiple agents, open source

MiroFish

A simple and universal swarm intelligence engine that can predict everything. Based on swarm intelligence algorithms, suitable for trend prediction and multi-agent collaborative decision-making scenarios.

7288
View repository
OthersPydantic AIPW
Pydantic, type safety, agent, Python

Pydantic AI

Based on Pydantic's Agent framework, type-safe AI application development

7287
View repository
OthersHello AgentsPW
tutorial, agent, Chinese, beginner

Hello Agents

Datawhale's Chinese tutorial on building agents from scratch, teaching you step-by-step from principles to practice in building AI agents

7290
View repository
OthersWhisperPW
Speech recognition, ASR, multilingual, OpenAI

Whisper

OpenAI's open-source general speech recognition model supports multilingual speech-to-text and translation, with extremely high accuracy

7185
View repository
OthersnanochatPW
LLM training, GPT, Karpathy, minimalist

nanochat

Karpathy's minimalist LLM training framework can complete the entire process from tokenization to pre-training, fine-tuning, evaluation, and inference with a single GPU node. For just $48, you can train a GPT-2-level model and interact with it through a ChatGPT-style web UI.

7190
View repository
Efficiency toolsLightpanda BrowserPW
browser, headless, automation, AI, Zig

Lightpanda Browser

A headless browser designed specifically for AI and automation, written in Zig, delivers extremely high performance. Ideal for building AI-driven web scraping, testing, and automation workflows.

7185
View repository
Efficiency toolsFinceptTerminal is a financial application integrating global data, CFA-grade algorithms, and AI strategies, which can assist with stock selection, valuation, portfolio planning, and morePW
AI finance, AI

FinceptTerminal is a financial application integrating global data, CFA-grade algorithms, and AI strategies, which can assist with stock selection, valuation, portfolio planning, and more

FinceptTerminal is a financial application that combines global data, CFA-grade algorithms, and AI strategies to support stock selection, valuation, and portfolio management. It can connect multiple data sources and has multiple built-in investment strategies. Users can analyze by dragging and dropping ready-made modules, simplifying the investment decision-making process

7188
View repository
AI and AgentsAReaLPW
Reinforcement learning, LLMs, reasoning, agents

AReaL

Ant Group's open-source LLM inference and agent reinforcement learning framework offers ultra-fast training and simplicity and flexibility, taking large model inference capabilities to the next level.

7087
View repository
Efficiency toolsMicrosoft and the National University of Singapore have developed a 3D/4D scene generation project called GenXD, which can handle both 3D and 4D scene generation simultaneouslyPW
3D/4D generation models

Microsoft and the National University of Singapore have developed a 3D/4D scene generation project called GenXD, which can handle both 3D and 4D scene generation simultaneously

GenXD is a 3D/4D scene generation project developed jointly by Microsoft and the National University of Singapore. It can simultaneously process both 3D and 4D scene generation. The generated videos follow camera trajectories, maintain consistent object shapes, support 3D format conversion, and are suitable for virtual reality, gaming, and architectural visualization. The model supports processing of real-world and synthetic datasets, providing both single- and multi-view 3D and 4D generation.

7086
View repository
OthersMiniCPMPW
On-device model, lightweight, mobile deployment, wall-facing intelligence

MiniCPM

Mianbi Intelligent's open-source on-device large model series offers 2B parameters comparable to 7B performance and supports mobile deployment

6982
View repository
Efficiency toolsSuryaPW
OCR, multilingual, document analysis, 90+ languages

Surya

Multilingual OCR and document analysis tools, supporting 90+ languages with extremely high accuracy

6983
View repository
Writing and ResearchSEO MachinePW
SEO, Claude Code, content generation

SEO Machine

An SEO long-form content workstation based on Claude Code automatically researches, writes, analyzes, and optimizes top-ranked blog content, suitable for any business scenario.

6982
View repository
OthersMobileAgentPW
agent, mobile, gui, multimodal

MobileAgent

Alibaba's X-PLUG mobile GUI Agent series supports multimodal visual perception and autonomous mobile interface operation, suitable for automated testing and mobile assistant scenarios. ⭐ 7.9k

6987
View repository
OthersAReaLPW
reinforcement-learning, llm, reasoning, training

AReaL

Huawei's inclusionAI open-source LLM reinforcement learning framework focuses on enhancing inference and agent capabilities, offering fast speed, flexibility, and easy scalability. ⭐ 4k

6990
View repository
Efficiency toolsThe latest 1B end-to-end OCR: LightOnOCR-2-1B, with a single H100 card delivering 5.7 pages per second, costing less than $0.01 per thousand pagesPW
OCR

The latest 1B end-to-end OCR: LightOnOCR-2-1B, with a single H100 card delivering 5.7 pages per second, costing less than $0.01 per thousand pages

The latest LightOnOCR-2-1B model performs excellently on OlmOCR-Bench, 1.7 times faster than OlmOCR, capable of handling complex documents and multilingual content. The H100 card handles 5.7 pages per second, costing less than $0.01 per thousand pages

6991
View repository
OthersMiroFishPW
swarm-intelligence, prediction, python, ai

MiroFish

A simple and universal swarm intelligence engine that can be used to predict everything. Based on swarm intelligence algorithms, it supports multiple prediction scenarios.

6989
View repository
AI and AgentsMiroFish - Swarm Intelligence Prediction EnginePW
swarm-intelligence, prediction, python

MiroFish - Swarm Intelligence Prediction Engine

A simple and universal swarm intelligence engine that uses Swarm Intelligence to predict everything. Implemented in Python, supports multiple prediction scenarios, and is easy to scale. Today, GitHub has 3200+ stars.

6881
View repository
Efficiency toolsChatGPT-Next-WebPW
ChatGPT, one-click deployment, multi-model, cross-platform

ChatGPT-Next-Web

One-click deployment of your own ChatGPT web application, supporting multiple models such as GPT-4 and Gemini Pro

6881
View repository
Efficiency toolsnotebooklm-pyPW
NotebookLM, Google, API, Python

notebooklm-py

Google NotebookLM is an unofficial Python API that programmably calls NotebookLM to generate podcasts, summaries, and other content

6883
View repository
Efficiency toolsA Google project that generates 4D scenes (dynamic 3D) from monocular video: CAT4DPW
3D/4D generation models

A Google project that generates 4D scenes (dynamic 3D) from monocular video: CAT4D

Google's CAT4D project can generate dynamic 3D scenes from monocular video, supports separate control of camera angle and timing, and offers three output modes. This project outperforms other models in reconstruction and is suitable for real or generated video, mainly used for cinematic effects. Currently, only Chrome 130+ browser is supported.

6782
View repository
Efficiency toolsA practical code review skill pack: code-review-expert, allowing AI to conduct comprehensive, structured review of code like a seasoned engineerPW
skills, AI

A practical code review skill pack: code-review-expert, allowing AI to conduct comprehensive, structured review of code like a seasoned engineer

code-review-expert is an open-source code review skill pack that simulates the perspective of a senior engineer, covering key dimensions such as architecture, security, and performance. It offers features such as SOLID principle checking, security scanning, performance optimization, error handling, and code cleanup, and features a seven-step review process and a P0-P3 grading system

6779
View repository
AI and AgentsOpenHandsPW
Software development, agents, bug fixing, project management

OpenHands

An AI-driven software development agent that can write code, fix bugs, and manage projects

6781
View repository
Efficiency toolsManimPW
Mathematical animation, teaching, Python, 3B1B

Manim

3Blue1Brown uses a Python engine for creating mathematical animations, which can be AI-assisted in generating teaching animations

6780
View repository