Importing an unsupported document

AIsuru supports importing various types of text documents, but there may be times when you want to import a format that isn't officially supported. This guide will help you handle importing different types of non-standard files, making sure your AI Agent can access and effectively use that information.

General procedure

If your document format isn't directly supported:

1. Extract the text or information from your file;

2. Paste the extracted information ->

   1. Into the instructions if you want the AI to always have everything available (and if the text is under 200,000 characters);

   2. Or into contents if you want to separate the information by topic ->

      1. Question -> enter a keyword;

      2. Question variants -> write specific questions for which that information is useful;

      3. Answer -> paste the text you extracted;

3. Explain how to read and use the data in your Agent's instructions.

Handling specific formats

Now that we've covered the general procedure, let's look at how to handle specific file formats.

Excel files

Simpler Excel files — and tables in general — can be converted to markdown format and inserted into a content or your Agent's instructions.

If your Excel file is more complex and you need higher accuracy in responses:

1. Identify different categories to group the data into;

2. Starting from your Excel file, extract the information and save it in separate JSON files (one per category) with a clear and simple structure that anyone could understand;

3. Copy the JSON code and paste it ->

   1. Into the instructions if the total doesn't exceed 200,000 characters

   2. Into separate contents (one per category, matching the files) ->

      1. Question -> enter a keyword;

      2. Question variants -> enter examples of questions the AI should be able to answer from that data;

      3. Answer -> paste the JSON code.

   3. Into separate functions (one per category, matching the files) -> you can also use functions to host your content. Just make your JSON file accessible and the Agent will retrieve the information when needed

4. In the instructions, tell the Agent where to find the information (in the instructions, contents, or functions), how to read the data (the JSON structure), and how to use it to respond to user requests.

Video

AIsuru doesn't officially support training from video files. If you want to make a video queryable:

1. Get a transcript of the video (use OpenAI's Whisper or use SharePoint for Teams meetings);

2. In the instructions, explain what the AI should do, paste the transcript.

✅ Done! The AI will now be able to answer questions about the video. Response quality will depend heavily on the transcript quality.

If you're using Microsoft SharePoint, you can also attach the video link (at the exact timestamp) to each answer:

1. Make sure to include the timestamps from the transcript in your Agent's instructions;

2. Copy the video link;

3. Customize and add these instructions:

1. VIDEO LINK:
 - The video is here: `...name.mp4`

2. TRANSCRIPT:
 - The transcript precedes each section with hh:mm:ss indicating the exact hour, minute, and second when something is said.
 - You MUST take the transcript, the EXACT timestamp, and the following rule into account when creating the link to a specific moment.

3. BUILDING THE LINK TO A SPECIFIC MOMENT:
 - If the timestamp indicates, for example, 5 minutes and 16 seconds, that means `5*60+16 = 316` seconds. You must then append this variable part to the link:
 `&nav=%7B%22playbackOptions%22%3A%7B%22startTimeInSeconds%22%3AXXX%7D%7D` where `XXX` must be replaced with the exact number of seconds.

### Example:
- Calculation: `5 minutes and 16 seconds = 5*60+16 = 316 seconds`
- Variable part: `&nav=%7B%22playbackOptions%22%3A%7B%22startTimeInSeconds%22%3A316%7D%7D`
- Full link: `...name.mp4&nav=%7B%22playbackOptions%22%3A%7B%22startTimeInSeconds%22%3A316%7D%7D`

These instructions were created for Microsoft SharePoint. If you're using YouTube or another service, make sure to modify the variable part of the link accordingly.

Images

To import information from images:

1. Use ChatGPT or Claude to get an accurate and complete description of the image;

2. Export the text and insert it ->

   1. Into the instructions -> if you want the Agent to always have the extracted information in mind;

   2. Into a content -> if you want to show the image on specific occasions (with certain keywords or questions):

      1. Question -> enter a question that the image answers (or a keyword);

      2. Question variants -> add other questions for which the image might be useful;

      3. Answer -> the image description;

      4. Media -> add your image.

Importing unsupported documents takes a bit more manual work, but it lets you integrate a wide range of information into your AI Agent. Choose the most appropriate method based on the type of document and the structure of the information it contains. Always verify that the imported information is accurate and easily accessible to the Agent, to ensure precise and relevant responses.