Description:
Open-source LLM-powered knowledge platform for enterprise document understanding, semantic retrieval and reasoning; turns raw documents into a queryable RAG, autonomous agent and self-maintaining wiki.
Data Science, AI and Machine Learning
Whirl
Description:
Whirl is a full-stack AI chat app built with Next.js and Convex. It features living documents, long-term memory, live web search, and installable skills, with a self-hosted option and a polished, zero-retention design.
FreeLLMAPI
Description:
FreeLLMAPI is a local router and MCP server that aggregates 34 free LLM providers behind a single OpenAI-compatible /v1 endpoint. It offers 7.4B tokens/month across 474 model families and 635 endpoints, automatically routing requests, refreshing catalogs, and encrypting keys while remaining compatible with OpenAI-like clients.
VoxCPM2
Description:
VoxCPM2 is an open-source tokenizer-free Text-to-Speech system enabling multilingual speech synthesis, voice design, and controllable voice cloning. Built on a diffusion autoregressive architecture, it delivers 48kHz studio-grade audio across 30 languages, with training, fine-tuning (SFT/LoRA), and production deployment options.
Breeze TTS 2
Description:
Breeze TTS 2 is an open-weight, real-time text-to-speech model with voice cloning, voice design, and voice direction. It targets expressive, low-latency synthesis (English and Chinese) using PyTorch, with downloadable model weights and self-hosted inference.
oh-my-hermes
Description:
oh-my-hermes (OMH) is an open-source plugin for Hermes Agent that adds planning, category-based routing, evidence-backed workflows, and a memory store, providing a transparent operating layer above Hermes to coordinate tasks.
Machine Learning From Scratch
Description:
Machine Learning From Scratch provides bare-bones NumPy implementations of classic ML models and algorithms, focused on accessibility and transparency from linear regression to deep learning.
PersonaPlex
Description:
PersonaPlex is a real-time, full-duplex speech-to-speech model that enables persona control via text-based role prompts and audio-based voice conditioning. Trained on synthetic and real conversations, it delivers natural, low-latency interactions with a consistent persona, built on Moshi architecture with NVIDIA model weights.
LongCat-Video
Description:
LongCat-Video is a 13.6B open-source video generation model enabling Text-to-Video, Image-to-Video, and Video Continuation with efficient, long-video generation at 720p/30fps using a unified architecture and multi-reward RLHF.
LongCat-Video-Avatar 1.5
Description:
An open-source framework for audio-driven human video generation. LongCat-Video-Avatar 1.5 improves lip-sync with Whisper-Large, enhances production stability for long videos, generalizes to stylized domains, and enables faster 8-step inference.