Desktop & tooling
LocalMind
Chat, document Q&A and computer vision, running entirely on your own machine.
- Role
- Sole author — application, backend and interface
- Period
- 2025 — 2026
- Status
- Live
The problem
The useful AI tools each want a subscription, an API key, and a copy of whatever you feed them. That is a poor trade for the ordinary cases — asking questions about a contract, pulling text out of a scan, counting objects in a photo — where the material is exactly the sort of thing you would rather not upload.
The approach
LocalMind is a desktop application that runs the whole stack on the machine it is installed on: chat and document Q&A against a local Ollama model, and vision through YOLOv8, EasyOCR and OpenCV. Documents are chunked into roughly 800-word segments and matched by keyword, so only the relevant passages are passed as context — which is what keeps a 100-page PDF answerable without a vector database or a cloud service.
What it does
Local chat with streaming
Conversational AI against any Ollama model, streamed token by token, with session history and markdown and code-block rendering.
Document Q&A over your own files
PDF, DOCX, TXT, MD and CSV. Load several at once and scope a question to a selection or ask across all of them.
Three offline vision modes
YOLOv8 object detection across 80+ classes with four selectable model sizes, EasyOCR text extraction, and OpenCV Haar-cascade face detection that needs no model download at all.
No key, no account, no upload
There is no API key to obtain and no service to sign into. Model weights download once; everything after that is local.