Composer, Shortcuts & Attachments

1. Message input

The field grows as you type. Its lower bar contains attachments, reasoning, RAG selection and context; tools and MCP are configured in Settings.

Press Send to start a response. During generation, the button lets you stop it. Stop requests cancellation: it does not undo an edit, command or web action that has already run.

2. Keyboard shortcuts

• Enter: send the message from the text field.
• Shift + Enter: insert a line break.
• Escape: close dialogs or menus that support this shortcut.
• ↑ / ↓: move through profile selector options.
• Ctrl / Cmd + V: paste text; image pasting depends on clipboard content and permissions.

3. Attachments

Use the paperclip, drag files into the chat or paste an image. Review attachments before sending and remove unwanted ones with their close button.

• Images: require a vision model; accepting a file locally does not ensure the provider accepts its format or size.
• PDF: text is extracted in the browser. Scans without text and complex layouts may produce incomplete results; there is no general OCR.
• Office documents: Word (DOCX), Excel (XLSX), PowerPoint (PPTX) and OpenDocument (ODT, ODS, ODP) are converted in the browser to text with headings, lists and tables; presentations are marked per slide and each spreadsheet is sent as a table of up to 500 rows. Up to 4 images per document are sent to the model if it supports vision. Macros are never run and legacy formats (DOC, XLS, PPT) are not supported: save them in the current format.
• Text and code: supported formats include TXT, Markdown, JSON, JavaScript, Python, HTML and CSS.
• CSV and TSV: sent as text for the model to analyse; check delimiters, units and results.

The first time you attach an office document, its reader library is downloaded from the ZeroChat site itself and cached; offline, it only works if it was downloaded before.

Attachment content is sent to the provider with the message and consumes context. To consult parts of several documents, use the RAG collection.

4. Message presentation, tables and documents

Responses support Markdown, tables with horizontal scrolling and CSV copy and download controls (RFC 4180), code blocks with copy and preview buttons, and normalised mathematical notation. HTML or SVG code blocks and tool cards that write files (such as write_file) include a Preview button that opens the document or markup in a new tab, inside an isolated frame: its scripts run, but without access to ZeroChat cookies, storage or local server. File modification tool cards (write_file and edit_file) also include a Diff button that opens the graphical diff viewer configured in Git (git difftool) in an independent OS window to compare the current file against HEAD (or inspect the last commit if there are no pending changes). If Git has no graphical viewer configured, it offers the first one installed (VS Code, Meld, KDiff3, Kompare or gVim) or, if there is none, to install webdiff in ZeroChat's local environment, and shows the git config --global commands before applying them. It only appears on the local server tools. Images displayed in the conversation (attachments, web pages, RAG, or tool screenshots) can be clicked to open at full size in a new tab. Reasoning exposed by the provider appears in collapsible blocks; tool calls have cards showing arguments, results and status.

Formatting does not validate content: check code, calculations and links before using them.