#jupyter_notebook #gemini #large_language_models #llm #openai #training #transformers

This project shows how to build and train a transformer language model from scratch in PyTorch. It uses the Pile dataset, tokenizes text with tiktoken, and stores tokens in HDF5 files for faster training. The code includes attention, MLP, transformer blocks, training, saving, and text generation. The benefit is that you can learn how LLMs work and train your own small or large model on a single GPU, then use it to generate text for your own tasks.

https://github.com/FareedKhan-dev/train-llm-from-scratch

#python #agentic_ai #agentic_workflow #agents #function_calling #llama_cpp #llamafile #llm #ollama #python #self_hosted #tool_calling

Forge is a Python tool that makes self-hosted LLM tool-calling more reliable. It helps local models handle multi-step tasks with guardrails, better context control, and support for Ollama, llama-server, Llamafile, and Anthropic. You can use it as a workflow runner, middleware, or proxy server with OpenAI-style clients. The benefit is fewer broken tool calls, better results on small models, and easier setup for agent apps, chat tools, and long-running sessions.

https://github.com/antoinezambelli/forge

#python #ai_agents #claude_code #cloud_security #cybersecurity #devsecops #ethical_hacking #incident_response #infosec #llm #malware_analysis #mcp #mitre_attack #nist_csf #osint #penetration_testing #red_team #security #security_automation #threat_hunting #threat_intelligence

Anthropic Cybersecurity Skills is a free, open-source library with 754 ready-made security skills for AI agents across 26 domains. It helps an AI act more like a senior security analyst by giving clear workflows, checks, and verification steps. It also maps each skill to major security frameworks like MITRE ATT&CK, NIST CSF 2.0, ATLAS, D3FEND, and NIST AI RMF. The benefit is faster, smarter, and more reliable security work with less guesswork.

https://github.com/mukul975/Anthropic-Cybersecurity-Skills

#python #audio #audio_tokenizer #llm #multimodal #text_to_speech #voice_cloning

MOSS-TTS is an open-source family of speech and sound models for natural, high-quality audio generation, including voice cloning, multi-speaker dialogue, real-time speech, and sound effects. It supports 31 languages in v1.5, better voice stability, and pause control, and it also offers a lightweight Nano version that can run on 4 CPU cores. The benefit to you is simple: you can create realistic speech or sound for apps, demos, or products with strong quality, flexible control, and multiple ways to run it.

https://github.com/OpenMOSS/MOSS-TTS

#python #ai #ai_companion #ai_vtuber #ai_waifu #chatbots #live2d #live2d_web #llm #neuro_sama #ollama

Open-LLM-VTuber is an offline AI companion with voice chat, vision, and a Live2D avatar that works on Windows, macOS, and Linux. It can run as a web app or desktop pet, supports many LLM, speech-to-text, and text-to-speech options, and is highly customizable, so you can build a personal assistant that fits your style and keep chats stored for later.

https://github.com/Open-LLM-VTuber/Open-LLM-VTuber

#python #agent #ai #anthropic #claude_code #compression #context_engineering #context_window #cursor #fastapi #langchain #llm #mcp #openai #prompt_engineering #proxy #python #rag #token_optimization #tokens #typescript

Headroom is a local tool for AI agents that shrinks prompts, logs, files, and chat history before sending them to an LLM, often cutting tokens by 60–95% while keeping the same answer quality. It can work as a library, proxy, MCP server, or agent wrapper, so you can save tokens, speed up workflows, and still recover the original content when needed.

https://github.com/chopratejas/headroom

#python #ai_agent #algorithmic_trading #backtesting #fintech #llm #mcp #multi_agent #python #quantitative_finance #trading

Vibe-Trading is an open-source trading research tool that turns plain English questions into market analysis, backtests, reports, and trading exports. It helps you test ideas, study your own trades, and run agent teams, so you can make faster, better trading decisions without building everything from scratch.

https://github.com/HKUDS/Vibe-Trading

#javascript #ai #anthropic #caveman #claude #claude_code #llm #meme #prompt_engineering #skill #tokens

This tool helps agents answer in caveman style: fewer words, same technical meaning, and often much lower token cost. You can choose levels like lite, full, ultra, or wenyan, and it also includes commands for shorter commit messages, quick PR reviews, stats, and compressing memory files; the benefit is faster replies, lower cost, and longer context for your work.

https://github.com/JuliusBrussee/caveman

#python #agent_infrastructure #ai_agent #ai_search #automation #bilibili #claude_code #cli #cursor #free_api #llm_tools #mcp #python #reddit_scraper #twitter_scraper #web_scraper #xiaohongshu #youtube_transcript

Agent Reach helps your AI agent get internet access fast, so it can read webpages, YouTube, Reddit, Twitter/X, GitHub, RSS, and more with simple commands. It is free, open source, privacy-focused, and works with many agents that can run terminal commands. This saves you time because you do not need to set up each site by hand, and your agent can start finding, reading, and summarizing online content for you in minutes.

https://github.com/Panniantong/Agent-Reach

#python #ai #chromadb #llm #mcp #memory #python

MemPalace is a local AI memory tool that saves your chats as exact text, then finds useful parts with search, so nothing leaves your machine unless you choose it. It is free, open source, MIT-licensed, and works with Python 3.9+, ChromaDB, and local tools like Claude Code. The main benefit is safer, private, and more reliable memory for your projects, with no required API calls or cloud fees.

https://github.com/MemPalace/mempalace