Allow Attaching Training Files to An Agent Without Assigning a Model
At the moment, attaching a training file to an agent is only possible if a model is assigned. Removing this restriction would allow training files to be added without assigning a model. The issue I'm encountering is that I need to include a table as an example in the system instruction for an agent, but the only way to do so is by adding it as a training file, which requires linking it to a specific model. I’d like to add the table as an example in the system instruction without being restricted to a particular model.
Log in to comment and vote
Comments3
Ngoc Nguyen
Oct 22, 2025
As of now, training files are injected directly into the model’s system instructions. This means they are limited by the context length of the chosen model. To ensure your training files work properly, you need to select the base model first, so the file size fits within that model’s context limit.
Azure Chipmunk
Oct 22, 2025
I appreciate what you are trying to do, but it would be better if you included an estimate of the amount of tokens associated with the training files instead of forcing me to choose a model. Perhaps you can even add a token count for the system instruction for the agent on the chat screen so that when the user selects the agent, it will show that the system instruction is already X number of tokens. It just seems like there must be a more flexible simple way to make this work than to require a model to be tied to the agent.
Azure Chipmunk
Nov 14, 2025
I’m bumping up against this limitation again and it is very frustrating. I want to create agents with system instructions that reference specific training files. However, tying the agent to a single AI model (not even multiple models) makes this a non-starter. So, now what I have to do is attach the same training files every time I use the agent. It’s maddening what a waste this is.