Independent Column

Open-source projects worth exploring

Runnable, documented projects worth studying and extending.

OthersNanochat - Karpathy's $100 ChatGPTPW
ChatGPT, Karpathy, minimalist, open source

Nanochat - Karpathy's $100 ChatGPT

Andrej Karpathy's latest project: creating the best ChatGPT alternative for $100, with minimalist design and refined code.

5770
View repository
OtherspromptfooPW
Prompt testing, red teaming, evaluation

promptfoo

Prompt/Agent/RAG testing and red team evaluation tools, supporting performance comparison and security scanning of mainstream models such as GPT, Claude, Gemini, Llama, and more

5774
View repository
OthersGuardrailsPW
Security guardrails, verification, anti-hallucination, LLM

Guardrails

Adds structured validation and security guardrails to LLM outputs to prevent hallucinations and formatting errors

5772
View repository
OthersQwen-AgentPW
AI, Agent, LLM, MCP, Python

Qwen-Agent

Tongyi Qianwen official agent framework, supporting Function Calling, MCP, code interpreter, RAG, Chrome extensions, and more, built on Qwen 3.0+.

5776
View repository
Efficiency toolsTabbyPW
Code completion, Copilot replacement, self-hosting, open source

Tabby

A self-hosted AI code assistant, an open-source alternative to GitHub Copilot, supporting local deployment

5773
View repository
OthersClaude CodePW
AI programming, terminal tools, code assistant, Anthropic

Claude Code

Anthropic's official AI programming assistant understands your codebase, executes daily tasks, interprets complex code, and handles Git workflows through natural language commands. A phenomenon-level tool with 73k+ stars.

5772
View repository
OthersGLM-4PW
Zhipu, bilingual, 128K, multimodal

GLM-4

Zhipu AI's open-source bilingual dialogue model supports 128K context and multimodal understanding

5773
View repository
Efficiency toolsSTORMPW
Long Writing, Automated Research, Stanford, Wikipedia

STORM

Stanford's open-source AI long-form writing system automatically researches and generates Wikipedia-level articles

5771
View repository
Efficiency toolsA high-fidelity single-image to 3D generation model: TripoSG, with 1.5 billion parameters, capable of handling complex image featuresPW
3D generation, modeling

A high-fidelity single-image to 3D generation model: TripoSG, with 1.5 billion parameters, capable of handling complex image features

TripoSG is a high-fidelity single-image to 3D generation model with 1.5 billion parameters, combining correction flow linear trajectory modeling with Transformer architecture, capable of stable modeling of 3D object shapes and processing photos, cartoons, and sketches

5775
View repository
OthersMiroFishPW
Multi-agent, prediction, swarm intelligence, simulation

MiroFish

A multi-agent swarm intelligence prediction engine simulates future trends by building a high-fidelity digital world. Upload seed materials and describe forecast needs in natural language to obtain detailed forecast reports and interactive digital sandboxes.

5775
View repository
Efficiency toolsClaude CodePW
Terminal, AI programming, Anthropic, command line

Claude Code

Anthropic's terminal AI programming tool collaborates with Claude directly on the command line

5776
View repository
OthersMuseVPW
Virtual humans, video generation, diffusion models, Tencent

MuseV

A virtual human video generation framework based on diffusion models, supporting unlimited video lengths

5773
View repository
Efficiency toolsLM StudioPW
Desktop application, native LLM, easy to use, downloadable and run

LM Studio

Discover, download, and run desktop applications of local LLMs on laptops

5670
View repository
Efficiency toolsFirecrawlPW
Web crawling, Markdown, data cleaning, LLM

Firecrawl

Converts web pages into clean Markdown data usable by LLMs, supporting crawling of entire websites

5668
View repository
OthersMem0PW
Memory, personalization, context, persistence

Mem0

The intelligent memory layer adds persistent personalized memories to AI applications, supporting both user-level and conversational-level memory

5670
View repository
Efficiency toolsOllamaPW
On-premises deployment, Llama, model inference, Go

Ollama

Run large language models like Llama 3, Mistral, and Gemma locally—simple and easy-to-use command-line tools

5671
View repository
OthersMiroFishPW
Swarm intelligence, prediction, AI

MiroFish

A simple and universal crowd intelligence prediction engine, based on Swarm Intelligence for predictive analysis of various events, this week GitHub reached 17k+ stars

5668
View repository
OthersWorldMonitorPW
Intelligence analysis, news aggregation, geopolitics, dashboard

WorldMonitor

Real-time global intelligence dashboard—AI-driven news aggregation, geopolitical monitoring, and infrastructure tracking, all with a unified situational awareness interface.

5671
View repository
OthersOpenVikingPW
agent, context-management, memory, bytedance

OpenViking

ByteDance's open-source AI Agent context database uses file system paradigms to uniformly manage the memory, resources, and skills required by agents, supporting layered context and self-evolution. 15k+ stars。

5668
View repository
Efficiency toolsImmichPW
Photo management, AI recognition, self-custody, privacy

Immich

A self-hosted Google Photos alternative that supports AI facial recognition and object search

5670
View repository
OthersFish SpeechPW
TTS, voice cloning, multilingual, low latency

Fish Speech

Open-source TTS model supports multilingual zero-shot voice cloning, with latency as low as 150ms

5669
View repository
OthersHVE CorePW
Microsoft, Copilot, Prompt engineering, enterprise level

HVE Core

Microsoft's enterprise-grade prompt engineering framework provides a constraint-driven AI workflow for GitHub Copilot. Built-in RPI (Research→Plan→Implement) methodology, dedicated agents, and JSON Schema validation, suitable for individuals to large teams.

5673
View repository
OthersHere's a hardcore [Claude Code full suite setup]—10 months of practical experience distilled into one plugin, allowing Claude Code to directly upgrade to "professional" levelPW
Learning and tutorials

Here's a hardcore [Claude Code full suite setup]—10 months of practical experience distilled into one plugin, allowing Claude Code to directly upgrade to "professional" level

This is a systematic Claude Code configuration plugin, summarized after 10 months of practical experience, aiming to improve programming efficiency. Configuration covers agents, skills, hooks, and more, supporting high output and multi-instance parallel use. Users can make secondary modifications as needed to avoid memory issues caused by too many tools, and dynamic balanced usage recommendations are provided

5672
View repository
Efficiency toolsLocalAIPW
On-premises deployment, API, offline, model inference

LocalAI

An open-source OpenAI API alternative supporting native LLM runs, image generation, and audio transcription, completely offline

5669
View repository