Chat with Files
Turn your documents into an answering machine.
Don't spend hours reading a 50-page PDF. Ask Dhito a question, and it will use a local Large Language Model to synthesize an answer directly from your files.
Finding the document is only half the job
Search hands you a file. If it is a 90-page contract, a dense research paper, or a policy document written to be precise rather than readable, you are then on your own with Cmd + F — guessing at the exact wording of a clause you have not read yet.
Document chat closes that gap. Once Dhito has found the file, you can ask it a question in plain language and get an answer drawn from the text itself, with a pointer back to the page it came from.
How local document Q&A works
The mechanism is retrieval-augmented generation, run entirely on your machine. Dhito splits the document into passages and embeds each one into the same semantic index it uses for search. When you ask a question, it retrieves the passages most relevant to what you asked and passes only those to a local language model as context.
That retrieval step matters for accuracy as much as for speed. The model is not answering from memory or from anything it absorbed during training — it is answering from the specific passages in front of it, which is why it can cite them.
Both halves run on your Apple Silicon: the embedding model that does the retrieving and the language model that writes the answer. The document is never uploaded, and the conversation is never logged to a server, because there is no server.
Why this matters for confidential documents
Pasting a document into a cloud assistant is a disclosure. Depending on the product and plan, the text may be retained, reviewed by humans, used to improve models, or processed by sub-processors in other jurisdictions — and for material under privilege, NDA, or a data processing agreement, that can be a problem regardless of what the retention policy says.
Running the model locally removes the question rather than answering it. If the file never leaves the device, there is no transfer to disclose, no processor to assess, and no retention window to reason about.
Instant Summarization
Need to know what a dense legal contract says? Select the file and ask 'What are the termination clauses?'. Dhito reads the document and provides a clear summary in seconds.
100% Private & Offline
Uploading sensitive documents to cloud AIs is a massive privacy risk. Dhito runs the LLM entirely on your Apple Silicon. Your files never leave your computer, keeping your data secure.
Cited Answers
Dhito doesn't just guess. When it answers your question, it provides exact citations back to the source document, so you can verify the information instantly.
Common questions
Which language model does Dhito use?
Dhito runs a local LLM on your Mac's Apple Silicon. Nothing is routed to an external API, so document chat keeps working with no network connection.
Can it answer questions across several documents at once?
No. A chat session is scoped to one file: Dhito builds a temporary index from that file alone, answers from its passages, and tears the index down when the session ends. That is the trade for citations you can trust — every answer points back into the document in front of you rather than to somewhere in a library you would then have to go and check. Picking the file is not restricted to search results: the chat button at the top of the window (⌘⇧C) starts a session on any file you choose from disk.
How accurate are the answers?
Answers are grounded in passages retrieved from your own files, which removes most of the room a model has to invent things. Every answer carries a citation back to the source page, so verification is a click rather than an act of faith. A local model is smaller than a frontier cloud model, so for open-ended reasoning it is the weaker tool — for "what does this document say", it is the right one.
Does chatting with a PDF require an internet connection?
No. Indexing, retrieval, and generation all happen on your Mac. Dhito functions identically offline.
What file types can I chat with?
PDF, Word, RTF, ODT, EPUB, plain text, Markdown, CSV and code files directly, plus the transcripts Dhito generates from audio and video — so you can ask questions about a recorded meeting the same way you would about a contract. Excel, PowerPoint, Pages, Keynote and Numbers are not indexed, so they cannot be chatted with either.