Agents abstractions deprecated after Haystack 1.0 - why?

[M-JULIANI]

Genuinely would love to hear some more context around deprecating the agents API in Haystack 2.0?

Came across an interesting webinar demo-ing some of that functionality and was looking forward to implementing it.

Thanks in advance,
Marco

[M-JULIANI]

@julian-risch would be great to hear your input here, you seem to be developing some of the agentic workflows. Is it a case of popping down to haystack 1.0 in order to use some of the agentic abstractions? If I do that what are some of the tradeoffs of not using 2.0?

Thanks in advance,
M

[mrm1001]

Hi @M-JULIANI thanks for opening this discussion! We are going to be adding more agentic features these coming weeks, we would love to hear what your use case is, i.e. what are you trying to build with it?

In the meantime, please have a look at the agentic features in 2.x that you can start using to create some pretty advanced pipelines, such as:

• loops (https://haystack.deepset.ai/tutorials/28_structured_output_with_loop)
• routing (https://haystack.deepset.ai/tutorials/36_building_fallbacks_with_conditional_routing)
• building a chat application with access to tools and access to past messages (https://haystack.deepset.ai/tutorials/40_building_chat_application_with_function_calling)

[M-JULIANI]

Great to hear more resources are coming @mrm1001 ! I have been looking at the resources you provided this week. Pretty helpful!

Put simply, I'm trying to build a system that is able to:

1. put together a plan
2. answer elements in the plan using a set of tools available to it, this includes gathering data from multiple sources.