Documents & artifacts

Every conversation has a document workspace. When you ask the model to "create a document", "write a report", "draft a file" or similar, it uses the built-in document tools; no MCP server configuration needed. The result appears in the Documents panel (the file icon in the chat header), and you can review, edit, download, copy and send your changes back to the model.

How it works

• The built-in tools create_document, update_document, read_document, list_documents and delete_document are available in every chat with a tool-capable model (catalog models tagged tools).
• You can disable them globally in chat settings → MCP → "Document tools (built-in)". The five tool schemas add a small number of input tokens to every request; disabling saves those tokens.
• Document types: markdown (formatted preview), code (syntax highlighting), HTML (sanitized preview), CSV (table preview) and plain text.
• Documents are saved in your browser (IndexedDB), scoped per conversation, and each create/update also attaches a downloadable file snapshot to the conversation message, so documents survive reloads and are included in conversation export/import.
• The panel updates live while the model works, and opens automatically when a document is created.

Collaborative editing

Open the panel, hit the pencil icon, edit, and save. "Send edit to model" attaches your edited document to the next message, so the model continues working on your version (it reads it back with read_document and revises with update_document).

Uploading data

CSV and other text files are sent to the model as their full content. Excel files (.xlsx/.xlsm) are parsed in your browser and sent as CSV text (all sheets, capped at 100,000 characters), so the model can work with spreadsheet data the same way. Legacy .xls is not supported (save as .xlsx); dates arrive as raw Excel serial numbers; other binary formats (.docx, .zip) are rejected with a hint to convert to text.

Limits

• Documents are capped at 100,000 characters; up to 50 documents per conversation (250,000 characters in total across all documents of one conversation).
• read_document returns at most 30,000 characters per call; large documents are read in chunks with offset/length.
• Document names: letters, digits, spaces, dashes, underscores and dots, max 64 characters, unique per conversation (case-insensitive).
• Deleting a conversation removes its documents too.
• If you edit a document in the panel while the model is also updating it, the last write wins (no merge).
• Two tabs open on the same conversation share the same IndexedDB store: the last write wins, and the other tab picks up changes when the conversation is reopened.

Zero Data Retention posture

Document content never touches ozeye infrastructure. It is stored only in your browser, and it travels to the model provider you chose only as ordinary tool messages while the model works on it; the same privacy posture as your chat history. Nothing is logged, cached or retained server-side.