Chat with Documents

The next step in information processing

Instead of relying on public datasets and general knowledge, "Chat with Documents" generates context-specific answers and analyses based on your trusted internal sources. Upload your documents and use these documents as a basis for answering questions in the chat!

Documents centrally published by an administrator do not need to be selected by you. For a relevant regular chat question, AI-Public can automatically search the central knowledge documents available for your organization or location. You can also find these sources under Knowledge Base → Organization → Documents (or under your own location). Administrators can read in Setting up and managing the Knowledge Base how to add and publish this knowledge.

Overcoming data limitations

When you ask questions to a language model, you depend on the dataset with which the model was trained. This is generally information gathered from the internet. Non-public sources are likely not included in this dataset. By using your documents as a source for the chat, you ensure that the model has the information you need to answer your questions.

Possibilities with your documents

You can ask questions about your documents such as listing the main points of a document or summarizing the document. You can also have specific analyses performed by the language model using your own dataset.

Disadvantages of document-based chatting

Uploading documents and processing them are extra steps you do not have to take if you can get a good answer without the context of specific information. It also takes longer to generate an answer because the necessary information must first be retrieved from the document before the request can be sent to the language model.

Checking source usage

The amount of source text used depends on the question, the document, and the chosen model. Check which sources have been fully, partially, or not read in the answer. Read the explanation about file management and source usage.

Select one or more documents

You can enable file mode by clicking the paperclip on the right side of the question bar. You can select up to 10 files to chat with.

You can also paste an image, such as a screenshot or a copied picture, straight into the question bar with Ctrl+V (Cmd+V on a Mac); the image is uploaded right away and attached to your question, just like when you drag a file onto the question bar. To put a screenshot straight on the clipboard, press Cmd+Ctrl+Shift+4 on a Mac (without Ctrl, the screenshot is saved as a file) or Windows key+Shift+S on Windows. If you copy text that also contains an image, such as cells from Excel or text from Word, the text is pasted as usual. In an assistant, pasting images only works if that assistant allows adding images.

Suitable language models

When you start chatting with documents, it is checked whether the language model is suitable for chatting with documents. If this is not the case, a suitable model from the current catalog is automatically selected.

You chat with these documents as long as file mode is enabled.

Text to document

With Text to document you can create documents from the chat. This mode is intended for two types of documents:

• Free document — AI-Public creates a formatted HTML document that you can continuously improve with follow-up prompts.
• Template — you choose a Word template with fixed input fields. AI-Public then fills in the corresponding form based on the chat.

Read the full explanation in Creating documents.

If no document templates are available, AI-Public starts directly with a free document. If templates are set up, you first choose between Free document and one of the available templates.

Creating a free document

For a free document, the chat is on the left and the document preview on the right. Each prompt creates a new version of the document. You can go back to previous versions via the version bar next to the preview.

In the settings, you can choose among others:

• document or presentation
• portrait or landscape
• logo left, center, right, or no logo
• logo size
• document style

The formatting uses the branding of the location or environment. If no custom branding is set, AI-Public falls back to the standard house style.

Use annotations to precisely indicate which part of the document should be changed. For each annotation, you type directly what should change, for example that a title, block, or table should be different. You can then send the annotations without an extra prompt.

You can also add files to the document prompt, for example a recording of a meeting, a PDF, or an existing Word document. The AI then uses that content as context for the new document.

Free documents can be exported as PDF. The PDF uses the same document orientation as the set preview.

Filling in a template

For a template, a form appears on the right. The AI fills fields as soon as information from the chat is available. You can then check and adjust the fields yourself before generating the document.

Examples:

• You upload a recording of a meeting and ask for meeting minutes. The AI fills in participants, agenda items, and action points.
• You choose a letter template and describe the recipient, subject, and message. The AI fills the form and then creates the Word document.
• You choose a report template and provide the main findings and recommendations.

After generating, you can view and download the Word document. The formatting comes from the Word template, so the document remains editable.

Check the form

Template fields can be filled automatically, but always check the form before generating the document. Especially dates, names, action points, and amounts often require a final human check.

Process per file

Besides chatting with documents, AI-Public also offers the possibility to apply a prompt separately to each document and receive individual answers. This feature is called Process per file.

This feature can be used in combination with "Chat with files."

Possible scenario

A practical example of using "Process per file":

1. You upload a reference document and enable it under Chat with files
2. You upload multiple documents that need to be analyzed and enable them under Process per file
3. You formulate a prompt, which is applied to all files individually

In this way, you can automatically analyze all documents based on the reference document.

Maximum number of files

There is a maximum of 30 files for the "Process per file" function.

Supported file types

AI-Public supports various file types for chatting with documents:

• PDF files ending in .pdf
• Word files ending in .docx
• CSV files ending in .csv
• JSON files ending in .json
• Text files ending in .txt
• Audio and video files with the extensions 'mp3', 'mp4', 'mpeg', 'mpga', 'm4a', 'wav' or 'webm'

Chatting with audio or video files

For chatting with audio or video files, AI-Public first transcribes the file via the configured transcription provider, for example OpenAI or European AI. The specific models come from the central model catalog.

For conversations, the transcription can also include time blocks and speaker labels when the chosen transcription model supports that. Then a suitable text model can be used to correct punctuation, spelling, speaker labels, and technical terms.

After that, the same procedure follows as extraction from PDF or Word documents.

Audio and video models have provider- and model-dependent limits for file size and duration. Long files may therefore be processed differently than short files. If processing fails, check the file status and try to process the file again or submit it in smaller parts.

Files you can download as examples

Sample business report
Sample project plan
Large history document

Processing and reusing files

Files you upload are first processed before AI-Public can use the content in chats, assistants, and workflows. If processing fails, the file receives an error status and you can re-upload it or reprocess it from File Management.

PDF, Word, and PowerPoint files are automatically processed into searchable content. Text, document structure, and where possible relevant visual information are included, for example from scans, filled forms, handwritten notes, circled or underlined choices, tables, and images. Large documents can be split into smaller parts during processing.

When a form or workflow requests a file, you can upload a new file or choose an existing file via the media manager. Files added via such a form are available to the assistant in that chat but are not automatically selected for regular chat questions.

Markdown files with the extension .md are also supported.