Local-first RAG, without the server
How LoveMyDocs runs embeddings and grounded chat entirely in the browser so you can ask private PDFs without uploading them.
On this page
Most “chat with your docs” products ship your files to a remote model. LoveMyDocs does the opposite: retrieval-augmented generation (RAG) stays local-first. Embeddings for search and the chat model for Ask your docs run in the browser tab — WebGPU when the machine supports it, WASM when it does not.
If you are comparing tools that ask you to upload PDFs, this is the architectural difference that matters.
What does local-first RAG mean here?
RAG usually means: retrieve relevant passages from a corpus, then ask a language model to answer using those passages. In cloud products, both the index and the model often live on someone else’s servers, so your documents have to leave the device.
In LoveMyDocs:
- You link a folder of Markdown (and drop in PDF, CSV, or TXT as needed).
- Text is chunked and embedded in the browser for semantic search.
- Ask your docs retrieves passages from that on-device index and generates a grounded answer in the tab, with citations back to your files.
There is no LoveMyDocs account that uploads the vault for indexing. The folder on disk remains the source of truth; the browser index is a rebuildable cache.
What runs in the tab?
Three layers stay on the device:
- Embeddings — vectors used to find passages related to your question
- Retrieval — hybrid search over your indexed chunks (and related graph signals where available)
- Chat — an on-device instruct model that answers from the retrieved context
Your confidential PDFs and notes are not sent to a remote “chat with PDFs” API as the primary path. That is why the product can honestly say: ask private documents on your computer; nothing is uploaded.
How is this different from uploading to a cloud model?
| Approach | Where files go | Typical trade-off |
|---|---|---|
| Cloud chat / upload | Remote service | Convenience; weaker control over confidential files |
| LoveMyDocs | Folder on your disk; compute in the browser | Stronger privacy; desktop browser + model download |
Cloud tools are not “wrong” for public manuals or marketing drafts. They are the wrong default for medical records, contracts, and personal archives. Local-first RAG is for the files that must not become someone else’s training set or log line.
What are the honest trade-offs?
Local-first is not free in user effort:
- First download of models takes time. Chat weights and the embedder are cached by the browser after you fetch them.
- Chat prefers Chromium-friendly hardware. WebGPU helps; without it, search can still work while chat is constrained.
- Desktop only for now. Phones and tablets are not supported. You need Chrome or Edge on a computer and a folder on disk.
- Context windows are finite. Grounded answers depend on retrieval quality and the passages that fit in context — which is why citations and jump-to-passage matter.
The upside is permanent: your vault never has to leave home for the product to work.
When should you use LoveMyDocs instead of a cloud PDF chat?
Use LoveMyDocs when the question is something like “can I ask this PDF without uploading it?” and the answer must be yes. Use a cloud tool when the document is already cleared for third-party processing and you need mobile sync or shared workspaces.
For product positioning and who the app is for, start with Welcome to LoveMyDocs. Then open the app on your computer and try Ask your docs on a folder you would never send to the cloud.