Yenisei AgentFeatures
Working with files
How the agent reads and edits files, which documents it can read, and when it asks for permission.
The agent reads files in the working folder, creates new ones, edits existing ones, and searches for files by name and for lines by the text inside them. Just describe the task in words:
- "Rewrite the introduction in article.md to make it shorter."
- "Find all mentions of the old address in every file and replace them with the new one."
- "Sort the photos into folders by year."
You can view the folder's files in the right panel.
Permissions
Below the input field there is a switch that determines what the agent is allowed to change.
- "Change nothing". The agent only reads files and replies.
- "Work in my folder". The agent changes files only inside the working folder.
- "Full disk access". The agent can work with any files on the computer.
Access switch below the input field
The mode for new conversations is set in the settings.
A file outside the working folder
If, in "Work in my folder" mode, the task requires touching a file outside the folder, the agent stops and asks for permission: a "Waiting for confirmation" window appears with the buttons "Allow once" and "Decline". Nothing happens until you answer. If you decline, the agent says so and suggests how to manage without that file.
The agent asks permission for a file outside the working folder
Attachments
To attach a file to a message, click "+" below the input field and choose "File", or drag the file into the app window.
Documents
The agent reads documents from the working folder and those attached to a message. Write, for example: "Summarize contract.pdf".
Supported formats: PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx), OpenDocument (.odt, .ods, .odp), RTF, text files, and source code. Documents are parsed right on your computer; no credits are charged for reading.
Scans
A PDF without a text layer, that is, one saved as images, cannot be read by the agent on its own. It will ask for permission to send the file for text recognition and will send it only after you agree. Recognition uses credits.
Images
- Ask about an image. Attach an image to a message and ask a question. If the image is already in the working folder, name it, and the agent will look at it, telling you so first.
- Generate. Ask, for example: "Draw a logo for a coffee shop in blue tones". The finished image is saved to the "Images" folder inside the working folder.
- Edit an existing image. Attach it and say what to change.
If the agent refused to draw, the app reports: "The model refused to draw: the content filter was triggered. Well-known characters, trademarks, and real people are not allowed; describe the image in your own words".
Limits
- Documents: up to 25 MB. Up to 30,000 characters of an attached document go into the message; if the document is longer, the app warns that only the beginning is shown.
- Scans for text recognition: up to 10 MB.
- Images: up to 10 MB and up to 4096 pixels on the long side, up to 20 per message. Formats: PNG, JPEG, WebP, and GIF.