Chat With Your Own Documents (Local RAG)
Import PDFs, text and Markdown. They are chunked, embedded on your phone and indexed in a local vector store — contracts and medical letters never touch a website.
Import your own documents and the assistant can quote them. The whole pipeline — reading, splitting, embedding, indexing, searching — runs on the phone.
How it works
A document is split into overlapping chunks, each chunk is turned into a vector by a small embedding model running locally (Nomic Embed v1.5, EmbeddingGemma 300M or E5 Small, depending on your phone), and the vectors go into a vector index inside the app's own SQLite database. When you ask something, your question is embedded the same way and the closest chunks are retrieved and put in front of the model.
This is retrieval-augmented generation, and doing it locally is what makes it interesting: the usual version of this feature involves uploading your documents to a vendor.
What it is for
The documents worth asking questions about are usually the ones you would not paste into a website. A tenancy agreement. A medical letter. A contract under NDA. Case files. Lecture notes and a textbook chapter before an exam. An appliance manual at the moment the appliance stops working. A pile of research papers.
What it will and will not do
It will find the relevant passages and answer from them, with the source to hand. It will not read a five-hundred-page book in one go — retrieval means it reads the parts that match, which is the point. Answers are as good as the retrieval, so a well-scoped question beats a vague one.
Keeping it organised
Group documents into folders and scope a conversation to one, or collect sources and notes together in a notebook. The Document Analyst agent exists to search this index and cite what it found.
Looking for something else? Every page on this site.