Skip to content
OpenWaggle

Attachments and voice input

Share a file or screenshot with the agent, or dictate a message and review it before sending.

Attach a file when the agent needs something from outside the project, such as a screenshot, sample data or a document. When the file is already in your repository, reference it instead, so the agent reads the current version. You can also dictate a message instead of typing it.

Attach files

Click + beside the message box to open its menu:

The composer's + menu open, with Attach files, Reference project file, Use a skill and Start Waggle (open full size)

  1. Choose Attach files and pick the files, or drag files onto the message box.
  2. Check the file chips above your draft. Click a chip’s X to remove a file.
  3. Say what the agent should do with the files, then send the message.

For example:

Compare this screenshot with the checkout page in our app.
List the differences before making any changes.

You can attach up to 5 files to a message. A document can be up to 8 MiB, and the documents in one message up to 20 MiB in total. An image can be up to 100 MiB, because the agent reduces it to the size the model accepts before the request is sent. Nothing is sent until you send the message. With a hosted model, the attachment content in the request goes to that provider.

You can also paste instead of picking a file. Press Ctrl+V, or Cmd+V on macOS, while an image is on the clipboard and OpenWaggle attaches it. Text on the clipboard is pasted as text. Screenshots work this way: press Win+Shift+S on Windows or Cmd+Shift+4 on macOS, then paste.

The rest of the + menu:

  • Reference project file picks a file from your project. Typing @ in the message box does the same.
  • Use a skill picks a skill, a set of instructions for one kind of task. Typing / opens the same list, together with commands and Waggle presets.
  • Start Waggle picks a saved Waggle preset, so two agents work on the message in turns. See Waggle mode.
  • Use MCP servers, under This Session, appears once the session has started. Clear it to stop this session’s MCP servers. See MCP.

What the agent receives

File typeWhat the agent gets
Text and documentsThe extracted text, for files such as .txt, .csv, .json, .xml, .html, .docx, .rtf and .odt.
PDFThe extracted text with its page structure, not the PDF itself.
ImageThe image, if the model accepts images, plus any text read from it by optical character recognition.

Images can be PNG, JPEG, WebP or GIF. A HEIC or HEIF photo from an iPhone is converted to JPEG first, so it reaches the model like any other image.

OpenWaggle saves attachments with the session, as their details and extracted text. It loads image and PDF data again when a request needs it.

Not every model can look at images. OpenWaggle goes by what the selected model says it supports. A model without image input gets only the extracted text, which says nothing useful about layout, colors or diagrams. For those tasks, choose a model that accepts images.

View images in the conversation

Images you or the agent share appear in the conversation. Click one to zoom, pan, or move through that message’s images. To browse every image in the session, open one from Session Summary and resources. The viewer can download a copy, or open the original file when it is still on your computer.

When the agent’s reply shows a local image, OpenWaggle can save its own copy with the session’s outputs, so the image stays after the original is deleted. Images in the agent’s replies lists the image formats and the folders OpenWaggle copies from.

Dictate a message

Click the microphone beside the message box to speak instead of typing. OpenWaggle turns your speech into text on your computer.

  1. Click the microphone.
  2. Speak while the waveform and timer show that it is recording.
  3. Click the square stop button or press Enter to stop. OpenWaggle writes out what you said.
  4. Read the text, fix names and code terms, and send it as usual.

To skip the review, click the send button while recording. OpenWaggle stops, writes out the text and sends it in one go.

Voice input needs macOS 13.4 or later on a Mac.

Where your voice goes

OpenWaggle uses the Whisper base model on your computer, with automatic language detection, and never sends the audio to another service. The first time, it needs an internet connection to download the model. It then keeps the model in the models/ folder of the app data folder and unloads it from memory after 5 minutes without use.

The text it writes is an ordinary message. When you send it to a hosted model, that provider gets the text.

When dictation fails

If transcription fails or OpenWaggle hears no speech, a message appears above the message box. Close it, or start a new recording to clear it.