ChunkCanvas

🌐 ChunkCanvas.cc is a document-processing GUI designed for RAG workflows. It allows you to parse PDFs, process images, transcribe audio/video, and extract data from Excel/CSV files using either local engines (Ollama, vLLM, Docling) or cloud providers (OpenRouter). From there, you can chunk content with configurable LangChain's RecursiveCharacterTextSplitter, edit parsed text and chunks directly in the UI, generate embeddings with your preferred provider, and ingest (text, embeddings and metadata) into the selected vector database of choice.

demo.webm

Document Support

Format	Built-in	Docling	Ollama	vLLM	OpenRouter
PDF (text)	✅	✅
PDF (vision)		✅	✅	✅	✅
Images			✅	✅	✅
Audio			✅	✅	✅
Video			✅	✅	✅
Excel / CSV	✅

Embeddings & Vector Databases

Embeddings Provider Voyage AI ☁️ Cohere ☁️ OpenRouter ☁️ Ollama 🖥️ vLLM 🖥️

Vector Database Pinecone ☁️ ChromaDB ☁️ 🖥️ MongoDB Atlas ☁️ FAISS 🖥️

---

Quick Start (Docker)

git clone https://github.com/Rayen023/chunkcanvas.git
cd chunkcanvas/

docker compose up -d

Service	URL
Frontend	http://localhost:3000
Docling Backend	http://localhost:8020/docs
FAISS Backend	http://localhost:8010/docs
Chroma Database	http://localhost:8030

---

Manual Setup

Frontend

npm install
npm run dev

Backend (FastAPI + uv)

cd backend
uv sync
uv run uvicorn app.faiss_server:app --reload --port 8010
# separate terminal for Docling:
uv run uvicorn app.docling_server:app --reload --port 8020

---

Local models

vLLM — run an OpenAI-compatible server:

vllm serve Qwen/Qwen3-VL-8B-Instruct-FP8 --port 8000

Ollama — install from ollama.com, browse available models, then start the server:

• Vision models
• Embedding models

ollama pull <model-name>