Files and knowledge

A model answers best when it can see your material. From the composer you can attach files to a single message, or point the AI at a whole knowledge base it can search and cite. A response style then controls how the reply reads.

app.stickyprompts.com 1 Attach files 2 Pick a knowledge base 3 Response style

Attach a file to a message

1. Select Attach

   In the composer, select Attach (1) and choose one or more files from your computer.

2. Say what to do with it

   The file is part of your message, so tell the model what you want from it:

   Here is last quarter's Northwind sales export. Which five stores grew
   fastest, and what do they have in common? Answer with a table.

3. Send

   The model reads the file together with your prompt. You can keep asking about it in the same chat.

Which file types a model can read depends on the model. Models that accept files carry the file_upload tag in the model selector; see Choosing a model.

Attach or knowledge base? Attach a file when it matters for this conversation only. When the same documents come up again and again, such as your policies, product sheets or help center, put them in a knowledge base once and pick it whenever you need it. Files you add to a project are included automatically in every chat in that project.

Use a knowledge base from the composer

A knowledge base is a searchable collection of your documents. When you pick one, the AI looks up the relevant passages for your question and cites the file each claim came from, right after the sentence it supports.

1. Select the book icon (2) in the composer.
2. Choose the knowledge bases to use, for example Northwind Help Center.
3. Ask your question as usual.

Knowledge bases accept PDF, DOCX, XLSX, CSV, TXT, Markdown, EML, JPG, PNG and WEBP files. Creating one, organising it in folders and sharing it with colleagues is covered in Knowledge bases.

Response styles

Style (3) in the composer picks a response style: how the reply is shaped, from the built-in Prompty and Chatty to styles you write yourself, such as a numbers-first style for leadership updates. The chat header shows the style in use, and you can switch between messages. Creating, copying and editing styles is covered in Response styles.

Sensitive information in files and prompts

Your workspace admins can switch on Sensitive Info Detection. It scans what you type and the data tools bring back, and can warn you, block the message, or redact values such as card numbers, IBANs and phone numbers before they reach the model. Redacted values arrive at the model as placeholders like [payment_data] or [phone], so the model can still work with the rest of your message.

Related pages