Upload file option in the UI for the agents

[neehar18]

I am trying to build an agent that would require for me to be able to upload files from the local filesystem.

The eventual goal is to build a prototype in the BeeAI playground with two of my custom agents which can:

• Agent 1 (file reader): Read any file of the following formats: json, csv and pdf. Requires a file to be uploaded to the agent.
• Agent 2 (rag agent): User should be able to ask questions about the contents of the file uploaded and get answers to it.

I have the rag agent built and working. To build the file reader agent (Agent 1), I'm missing the file uploading ability in the UI. I also didn't notice this option for the other out of the box agents. Is this something that is available in the existing UI or not? If not, are there any future plans to get this in? Also, can you point me to the relevant place where maybe I can implement that feature myself.

Below is the demo of my rag_agent in conjunction with literature_review:

[jenna-winkler]

Hi @neehar18 it’s great to hear this is a feature you’re looking for because it’s in development and will be released soon! #622

Update on ETA? @tomkis @PetrBulanek @jezekra1

[jezekra1]

There is an open PR draft on both backend and frontend, backend should be finished by end of this week:

#769
#783

There might be some changes required for ACP or additional work to make this compatible for both self-registered and managed agents

[jezekra1]

RAG is not yet part of this - it is a separate complex topic for another iteration

[neehar18]

Thanks @jenna-winkler and @jezekra1 for the quick response. I look forward to this feature. Let's keep this discussion open for the follow up messages when the feature gets available.

[neehar18]

Also, I had a question about the design of the agents. All of these default agents (except the chat agent) are hands-off type. Which is one and done. One query, one response. No scope for follow up queries. Also, the playground feature by design seems to only support hands-off type agents at the moment. For my use-case, I would love to have the option to ask follow up questions, not just one and done.

For example: In the above screenshot (literature_review + my custom rag_agent), I would like to have the option to ask further follow up questions about Alzheimer's. But this only lets me ask one.

Is this feature available and I just haven't discovered yet? Or is it a limitation with the platform at the moment?

[tomkis]

Sequential workflow just sequentially calls the agent so that's why we are limited on handsoff - to clearly understand where the input and output is.

To get this working we would need some kind of supervisor workflow, where the supervisor would be of a type chat. This workflow is unfortunately not implemented yet.