The current implementation of image attachment is a bit clumpy. The use case is not very clear and maybe not desirable in some circumstances:
- when used with a project level agent/service, it sends all images in a project, which is quite dangerous as it may saturate the context quite quickly and maybe most importantly exhaust the account credit
- it seems we do not send images manually added to the context
In the case of agenting modes (Agent or Interactive) this issue may be less relevant because we can request the agent to read it, but it is still common to attach an image we want the AI to process.
The other aspect to consider is if we want to keep the image in the conversation memory as it may uselessly consume a lot of tokens.
@jGauravGupta, feedback?
The current implementation of image attachment is a bit clumpy. The use case is not very clear and maybe not desirable in some circumstances:
In the case of agenting modes (Agent or Interactive) this issue may be less relevant because we can request the agent to read it, but it is still common to attach an image we want the AI to process.
The other aspect to consider is if we want to keep the image in the conversation memory as it may uselessly consume a lot of tokens.
@jGauravGupta, feedback?