Voice and attachments

Typing is not the only way to brief the agent. On Pro and Max plans you can speak your message, and on every plan you can attach files, screenshots and images that the agent reads directly.

Voice input

Voice input is available on the Pro and Max plans.

1. Press the microphone button in the message box.
2. Speak. Take your time: recordings can be up to 10 minutes long.
3. Press the microphone button again to stop.

Windlass transcribes the recording and inserts the text into the message box. Nothing is sent to the agent yet: read the transcription, fix any names or numbers that came out wrong, add a file reference with @ if you like, and then send it like any other message.

When voice works best

• Long briefs. Explaining a feature, a bug or a data problem out loud is often faster than typing it, and you naturally include context you might otherwise skip.
• Thinking aloud. Describe what you have tried and what you expect. The agent handles a rambling description well, and you can trim the transcription before sending.
• Hands busy. Review something on another screen while dictating the next task.

Tips for good transcriptions

• Speak at a normal pace, in a quiet room if possible.
• Read the result before sending. Product names, variable names and numbers are the most common things to correct.
• For exact identifiers, such as a file path or a command, type them into the message after transcribing or reference the file with @.

Attachments

Attach a file to your message and send it. Windlass saves the file in the current project's Attachments folder, and the agent can open it by path:

/workspace/Project 1/Attachments/logs.txt

Attachments are ordinary files in your project, so they remain available in later sessions of the same project, show up on the Files page and can be referenced with @ in the message box.

What to attach:

• Data files, documents and logs the agent should read, analyze or transform.
• Code you want reviewed or changed when it is not already in the project.
• Images and screenshots, described below.

Images and screenshots

When you attach an image, Windlass automatically runs text recognition and saves the result as a text file next to the image, ending in .ocr.txt. The agent reads this file to get the text out of the image: an error dialog, a stack trace, a table in a slide, a page of notes.

This makes screenshots a practical way to hand over information. Rather than retyping an error, screenshot it, attach it and say "fix this". The agent opens the OCR text, reads the message and gets to work.

Size limits

Attachments count toward your storage. If a file is larger than the upload limit, split it, compress it or ask the agent to fetch the data another way, for example from a URL or through a connector.

Tips

• Attach only what is relevant. Every file the agent reads is processed as input tokens and consumes allowance. Point it at the file it needs instead of attaching a whole folder.
• Reuse uploads. Upload reference material once through the Files page, then reference it with @ across sessions instead of re-attaching it.
• Combine voice and files. Attach the file, then dictate what you want done with it.
• Keep secrets out. Do not attach files containing passwords, private keys or API tokens. See Security and privacy.