SamirHann/fitcoach-ai

FitCoach AI is a local multi-agent system for bodybuilding/strength training that answers from user-provided documents (RAG), runs fitness calculations, performs focused web searches, and handles natural conversation. Intended for coaches, athletes, researchers or developers who need sourced guidance and personalized metrics. Key features: - Router (LangGraph + Mistral via Ollama) dispatches messages to RAG, Tools or Chat agents with keyword fallback; routing adds ~5–10s latency on CPU. - RAG: PDF/TXT ingestion into ChromaDB (500-token chunks, overlap 50), embeddings with all-MiniLM-L6-v2, top-3 retrieval, mandatory citations; citation integrity enforced by _fix_citations(). - Tools: Tavily Search integration (API key), fitness calculators (1RM Epley, TDEE Harris‑Benedict, macros), regex param parsing, LLM-driven tool selection. - Chat: conversational agent using Mistral, short natural replies without document citations. - 3-turn sliding memory injected into prompts. - Dockerized setup; Ollama, ChromaDB and CLI entrypoint; Tavily API required; ~4 GB RAM minimum.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.