|
<< Click to Display Table of Contents >> AI Agents FAQs |
AI Agents are a powerful tool for automating complex tasks within your processes, enabling you to handle large amounts of data, generate detailed responses, and streamline workflows. To help you get the most out of this feature, we have compiled answers to some frequently asked questions. Whether you are wondering about input limits, file handling, or how to optimize performance, these FAQs cover the essential aspects of working with AI Agents.
AI Agent Technical Limits
Q: How many variables can an AI Agent have?
A: There is no hard limit, but it's recommended to keep the number manageable for clarity and performance.
Q: What is the character limit for an AI Agent's input prompt?
A: Up to 500,000 characters, allowing long and complex inputs including full documents.
Q: What is the maximum length of the AI Agent's response?
A: Up to approximately 16,384 characters (based on the GPT-4o token limit).
Q: How many files can I upload to an AI Agent?
A: There’s no fixed number, but the total content must fit within the 500,000-character input limit.
Q: What’s the maximum file size for uploads?
A: The total size of uploaded files must be within the input limit of approximately 500,000 characters. Keep in mind that file formats like .docx, .xlsx, or .pdf include metadata that reduces the number of characters that can be processed compared to plain text files. Additionally, large or high-resolution images in documents may reduce the amount of text that can be handled.
Q: What should I consider when uploading large files or images?
A: For tasks involving large files or images, it is recommended to use AI tasks rather than triggering these processes via buttons. Handling large or complex files can take time and may affect user experience. To ensure smooth performance, use buttons for smaller, more specific tasks.
Q: Can I run multiple AI Agents asynchronously in sequence?
A: If your process involves running several AI Agents asynchronously, consider whether the tasks can be executed in parallel instead of sequentially. Running tasks in parallel can help improve performance.
Q: When does performance degrade due to input size?
The total size of uploaded files must be within the input limit of approximately 500,000 characters. Keep in mind that file formats like .docx, .xlsx, or .pdf include metadata that reduces the number of characters that can be processed compared to plain text files. Additionally, large or high-resolution images in documents may reduce the amount of text that can be handled.
Q: How many AI Agents can I use in a single process?
A: The number of AI Agents you can use in a process without performance loss depends on the complexity of each task, the size of the input/output data, and the available system resources. It is recommended to monitor system performance and limit the number of AI Agents accordingly.
Q: GPT-4o vs GPT-3.5—what's the difference?
A: GPT-4o is a more advanced model, offering higher accuracy, better context handling, and improved reasoning for complex tasks. GPT-3.5, while faster and suitable for simpler tasks, is less effective in detailed or nuanced scenarios. If you are creating a new AI Agent, we recommend using GPT-4o. There are no cost differences between GPT-4o and GPT-3.5. Bear in mind that to use files as inputs for your AI Agent, you must use the GPT-4o model.
Q: Is there a time limit for AI Agent execution?
A: Yes, an AI Agent has a maximum execution time of 240 seconds (4 minutes). If the AI Agent takes longer than this to complete its task, a timeout will occur. This limit applies regardless of the number of AI Actions integrated or the complexity of the configuration. It is important to consider this when designing your AI Agent, especially if it involves complex prompts, multiple files, or integrations.
Date Outputs
Q: When an AI Agent output is a date, how is it displayed if it's free text?
A: In Bizagi, the system prompt specifies that dates must be returned in UTC format to ensure consistency.
Q: If the date output is mapped to a Bizagi date field, will it use that field’s format?
A: Yes. When mapped, the output will adopt the format of the target field. If the field is of type date-time, the time component will be included.
Number Outputs
Q: When a number is returned as free text, what format does it follow?
A: By default, the system prompt includes rules about decimal separators and number formatting.
Q: If the number is mapped to a numeric field in Bizagi, will it adopt that field’s format?
A: Yes, mapped outputs will adopt the formatting defined in the Bizagi number field.
Q: Do I need to specify number or currency formats in the prompt?
A: Normally no, since the system prompt defines decimal and thousands separators. However, for currencies or specific formatting, it's safe to specify this explicitly in the prompt to avoid ambiguity.
Data Type Mismatches
Q: What happens if an AI Agent’s prompt maps to a field of a specific type (e.g., Date) but the response is of a different type (e.g., Number)?
A: This will trigger an error during execution. The response must match the expected data type of the mapped field.
Output Variables – Collections
Q: How many output variables can I define in a collection-type output?
A: Each item in the collection can have up to 100 properties (OpenAI model limit).
Q: Can I use nested structures inside collections?
A: Yes, OpenAI supports up to 5 levels of nesting, but Bizagi Studio may restrict this to 1 level.
Q: Is there a size limit for the entire schema?
A: Yes, the schema must not exceed 15,000 characters, including property names, enums, etc.
Practical Recommendation: As a best practice, divide these limits by half to stay safe:
Max 50 properties per object
Max 2 levels of nesting
Max 7,000 characters per schema
Output Variables – Normal Fields (Non-Collection)
Q: How many “normal” (non-collection) output variables can I define?
A: You can define up to 100 properties per object, including both normal and collection outputs. If you use both, they count toward the same 100-property limit.
Best Practice: Again, consider aiming for 50 variables or fewer for performance and clarity.
Last Updated 5/12/2025 4:13:35 PM