Resolving Workflow Timeout Errors

Increase the timeout, or set it to None

w = MyWorkflow(timeout=None)

thanks, got it!

Hey Logan!
Another question plsss!
In the multi-agent example, you define :

class AgentConfig(BaseModel):
"""Used to configure an agent."""
model_config = ConfigDict(arbitrary_types_allowed=True)
name: str
description: str
system_prompt: str | None = None
tools: list[BaseTool] | None = None
tools_requiring_human_confirmation: list[str] = Field(default_factory=list)
# llm: LLM

How can I use ReAct agent instead of this?

look at the code, probably you have some code like:
agent = AgentConfig ...
replace by
agent = ReActAgent(
llm=OpenAI(model="gpt-4o-mini"), tools=tools, timeout=120, verbose=True
) (from this example)
https://docs.llamaindex.ai/en/stable/examples/workflow/react_agent/#run-the-workflow

@Logan M
Hi again,
would u pls help me to increase the 1024 length of Tool description in Workflow? No way to increase the length at all?
WorkflowRuntimeError: Error in step 'speak_with_agent': Tool description exceeds maximum length of 1024 characters. Please shorten your description or move it to the prompt.

I need it specifically for SQLQueryEngineTool in which I provide the schema of the DB. It usually goes too long to provide clear view of the schema

@Alwiiiiiiiin this is a limitation of openai's API, they don't allow longer descriptions
https://community.openai.com/t/function-call-description-max-length/529902/3

Oh😑,
U think put it in the LLM system message would be the solution?

Yea that could help!

@Gianluca
I still have difficulties to replace the regular agent with the ReAct agent. can you please help me with the workflow code that I current have (working fine with regular agent)

@Alwiiiiiiiin instead of doing llm.chat_with_tools() -> llm.get_tool_calls_from_response(), you'll either need to replace that logic entirely with just a react agent, or use the lower level functions to do react more from-scratch

There's a lower level example of doing react here
https://docs.llamaindex.ai/en/stable/examples/workflow/react_agent/

Note that if you are using react, its probably because you are using open-source llms? And open source llms really such at being agents for the most part, just an fyi

Thanks for your info! I will have a look at it.

The main issue I'm encountering with this workflow is the Orchestrator's limitations. Despite providing a detailed prompt with a list of agents and their descriptions, it struggles to handle queries that involve multiple sub-queries. The workflow's performance is highly sensitive to the Orchestrator prompt, and even shortening it sometimes disrupts the flow.

Compared to a single ReAct agent that connects to different tools, this multi-agent framework often fails due to the Orchestrator's weakness in task handling.

Do you have any suggestions for improving this?

Feel free to just create a tool so that agents can handoff directly

rather than routing through an orchestrator