Attachments
You can add images, PDFs, text files, or code to a message β via the π button, drag-and-drop, or paste.
How each type is handledβ
- Images are shown as thumbnails and sent to the model as base64 multimodal content on image-capable providers (OpenAI / OpenAI-Codex / OpenRouter / Claude / custom). If a backend or model can't accept multimodal content, it degrades gracefully to a normal stream error you can retry.
- Text and code files are extracted and inlined into the prompt as labeled fenced blocks.
- PDFs are extracted to text and inlined. Parsing uses pdf.js loaded on demand from a CDN only when you actually attach a PDF β it is never bundled into the plugin, so it adds nothing to the install size.
Chips and limitsβ
Attached files appear as a chip row above the textarea (image thumbnail or file icon + name + size, with a spinner while a PDF is read). Send is disabled while any attachment is still loading. Per-file size guards apply, with friendly errors.