Uploading & analyzing files

The agent can read documents, look at images, and transcribe audio — all on your own machine. Nothing you upload leaves your computer: documents are parsed locally, image analysis runs through your own Ollama, and audio transcription uses the same local Whisper model as voice mode.

The + button

Next to the chat input is a + button. Attach one or more files, add a question (or don't — "tell me what this contains" is the default), and send. What happens depends on the file type:

• Documents — PDF, Word (.docx), Excel (.xlsx), and any plain-text format are extracted to text and read. Ask for a summary, specific numbers, action items, anything.
• Images — screenshots, photos, scans. The agent answers questions about what's in them ("what does this chart show?", "read the text in this screenshot").
• Audio — voice notes, call and meeting recordings (mp3, wav, m4a, and most other formats). Transcribed locally, then the agent can summarize, extract action items, or act on what was said.

Pointing it at folders

You don't have to upload things one at a time. Give the agent a folder path in plain language — "look through C:\Users\me\Documents\Invoices and list everything from March" — and it can list, search, and read files anywhere on your machine. Anything that would change your files (move, rename, delete) always comes back as a proposal you confirm first, and deletes go to the recycle bin, never permanent.

The files drawer

The 🗂 button in the header opens a drawer showing everything in one place: files you've uploaded, images the agent has generated, and documents it has produced. Click anything to open it.

If image analysis says no vision model is installed

Looking at images requires an Ollama model that accepts them. The agent auto-detects one if you have it; if not, it will tell you exactly what to run (typically ollama pull llava). Document reading and audio transcription have no such requirement — they work out of the box.