AI Agents and Coding Assistants
The largest cluster on today's list is tools that let agents write, run, or audit code with minimal human intervention.
- CodebuffAI/freebuff - A free, open-source coding agent. (+141)
- chaitanyagiri/munder-difflin - A local multi-agent harness for running several agents in parallel. (+181)
- HKUDS/CLI-Anything - Makes any CLI tool accessible to AI agents without extra wrappers. (+384)
- kenn-io/agentsview - Local-first session search and token analytics for Claude Code, Codex, and 20-plus other coding agents. (+35)
- 0xSero/ai-data-extraction - Extracts personal data history from Cursor, Codex, Claude Code, Windsurf, and Trae. (+51)
- cursor/plugins - The official Cursor plugin specification and first-party plugins. (+144)
LLM Training, Inference, and Fine-Tuning
Builders continue to push LLM workloads onto consumer hardware.
- unslothai/unsloth - Local UI to run and fine-tune LLMs and diffusion models including Qwen3, DeepSeek-V4, and FLUX. (+572)
- MakazhanAlpamys/Soup - Fine-tune LLMs from a single YAML file; layer streaming fits an 8B model onto a 4 GB laptop GPU. (+443)
- THUDM/slime - A post-training framework for reinforcement learning scaling of LLMs. (+31)
- jundot/omlx - LLM inference server with continuous batching and SSD caching for Apple Silicon, managed from the menu bar. (+60)
- AlexsJones/llmfit - A single command to benchmark hundreds of models and providers against your local hardware. (+187)
- mostlygeek/llama-swap - Reliable model swapping for any llama.cpp or vLLM-compatible server. (+17)
- ollama/ollama - Get up and running with Kimi-K2.6, GLM-5.2, DeepSeek, Qwen, and other models locally. (+127)
- google-research/timesfm - Google Research's pretrained foundation model for time-series forecasting. (+109)
AI-Powered Video and Design Generation
Programmatic media creation is attracting steady attention.
- OpenCut-app/OpenCut - An open-source alternative to CapCut for video editing. (+150)
- remotion-dev/remotion - Make videos programmatically with React. (+87)
- harry0703/MoneyPrinterTurbo - Generates HD short videos from a topic or keyword using an automated AI workflow. (+494)
- ZSeven-W/openpencil - An open-source AI-native vector design tool with concurrent agent teams and design-as-code support. (+138)
Security and OSINT
Security tooling continues to attract contributors.
- usestrix/strix - Open-source AI penetration testing tool that finds and suggests fixes for app vulnerabilities. (+856)
- smicallef/spiderfoot - Automates OSINT for threat intelligence and attack surface mapping. (+54)
- megadose/holehe - Checks whether an email address is registered across sites like Twitter and Instagram. (+231)
- EFForg/rayhunter - Detects cell site simulators on an Orbic mobile hotspot. (+26)
Edge and Embedded AI
Small models for constrained environments are gaining ground.
- cactus-compute/needle - A 14 MB foundation model targeting phones, wearables, smart home devices, and robots. (+443)
- newton-physics/newton - GPU-accelerated physics simulation built on NVIDIA Warp, aimed at robotics researchers. (+35)
- mayocream/koharu - ML-powered manga translator written in Rust. (+34)
Developer Workflow and Infrastructure
Everyday tooling that speeds up development without requiring AI.
- cordiverse/cordis - A meta-framework for spatiotemporal composability in TypeScript applications. (+720)
- dmtrKovalenko/fff - Fast and accurate file search SDK for AI agents, Neovim, and multiple runtimes. (+30)
- atuinsh/atuin - Synced, searchable shell history with statistics. (+40)
- sxyazi/yazi - Blazing fast terminal file manager built on async I/O. (+53)
- Automattic/harper - Offline, privacy-first grammar checker powered by Rust. (+40)
- bytecodealliance/wasmtime - A lightweight, standards-compliant WebAssembly runtime. (+8)
- public-apis/public-apis - A collective list of free APIs for rapid prototyping. (+1588)
- yt-dlp/yt-dlp - A feature-rich command-line audio and video downloader. (+216)
Today's trending list makes one pattern hard to miss: the developer toolchain itself is being rebuilt around agents, with projects targeting every layer from hardware-level inference to session analytics. At the same time, strong numbers on fine-tuning utilities like Soup and Unsloth suggest that founders are not just consuming AI, they are training and customizing models on their own machines.