Limitations¶

General¶

• You will get the best results for questions asked in English
• Always have in mind:

  AI-generated answer may be incorrect. Please consider verifying.

• Avoid selecting larger time ranges, because the data context will get too big
  If an error message is displayed, zoom in and try again
  
  The token limit for the context window in total is 128.000. But this is consumed by different sources like chat history, skill output, agent instructions, llm output etc.

Agents and skills¶

• The cycle time of a user question calling agents and skills is limited to 60 seconds in total.
• The cycle time of each requested skill within an agent is limited to 30 seconds.
• The number of agents per tenant is limited to 10.
• The number of skills per tenant is limited to 50.
• When a question is asked, a maximum of 10 skills are requested.
• The output of a skill is currently limited to 512 KB.
• Agents cannot be chained.
• VFC compute hours are used when a skill is triggered.
• Instructions of agents have to be guard-railed according to the use case.
• It is essential that custom agents and skills are tested properly.
• We recommend testing individual skills first and then proceeding to test their interaction in the context of an agent.
• We recommend to test based on typical user questions.

Files used in IDL¶

• The agent Knowledge Base or the skill findInDocuments utilizes IDL resources. Ensure sufficient IDL capacity to guarantee smooth operation.

  • If your tenant is using Resource Access Management, create a policy to the default copilot folder, that users can upload files directly from the chat window.
  • If the policy is missing, you will see a message like this while upload.
    
    
  • For more information on policies, refer to Resource Access Management.

• When an administrator syncs a file, the file, its content, or generated answers will be available to every user of the Production Copilot, even if they are not the owner of the file or have access to it in the IDL Manager.

• The files which are synced are sent to the LLM. Safeguard your privacy by refraining from sharing sensitive data.
• You will get the best results for questions asked in English.
• When you upload your own documents, note the following:
  • Supported file formats are pdf, json, txt, or csv.
  • Do not use umlauts, spaces, or special characters in the file name.
  • Check before upload if the text can be selected with the cursor.
    
  • Currently, pictures in documents are not analyzed.
    
  • The same applies to documents where the text cannot be selected with the cursor (e.g., scans).
  • For a question, relevant files are ranked by their relevance. A maximum of 4 different files will be returned in the answer.
  • Asking for a summary of a document is currently not supported.