ChunkCanvas

ChunkCanvas is a personal project, a document-processing GUI built for RAG workflows. It parses PDFs (text and vision-based), processes images, transcribes audio and video, and extracts data from Excel/CSV files. Parsing is handled through local engines Ollama or vLLM with vision-language models, or vLLM serving ibm-granite/granite-docling-258M in combination with Docling for structured PDF parsing. OpenRouter is available as a cloud fallback. Parsed content is chunked using LangChain's RecursiveCharacterTextSplitter with configurable parameters (chunk size, overlap and separators), and users can edit and correct both parsed text and chunks directly in the UI. Embeddings are generated with a chosen provider (local models via vLLM or Ollama, or cloud APIs via OpenRouter, Voyage AI and Cohere) and ingested alongside text and metadata into a selected vector database (ChromaDB, FAISS, Pinecone or MongoDB). The frontend is built with Next.js and the backend with FastAPI.