Agents Example: Multi-Agent via Tools/Handoff + Multiple Input & Output Guardrails + Streaming

API
1

spent the weekend exploring the new agents sdk — built a simple multi-llm call workflow and shared my first impressions the code example put together guardrails & streaming too!

This image is a flowchart illustrating a multi-agent workflow for handling a user's query, which involves query routing, tool calls, input/output guardrails, and error handling. (Captioned by AI)
This image is a flowchart illustrating a multi-agent workflow for handling a user's query, which involves query routing, tool calls, input/output guardrails, and error handling. (Captioned by AI)1498×658 63.8 KB

the code example puts together guardrails & streaming too!

thoughts:

  1. “agents as tools” vs “handoffs”
  2. the built-in traces ui is outstanding
  3. guardrails and streaming outputs: room for improvement?
  4. sometimes handoffs don’t stick
  5. potential latency bottlenecks
  6. the RECOMMENDED_PROMPT_PREFIX feels … fragile

more here: Gabriel Chua - OpenAI Agents SDK: First Thoughts

code here: GitHub - gabrielchua/oai-agents-example

7 Likes
2

Nice! Thanks for sharing with the community.

Any big obstacles you faced putting it together?

3

was initially quite confused with “handsoffs” and “agents-as-tools”, along with streaming the function calls!

for the latter, i basically logged everything and then read it line by line :stuck_out_tongue:

1 Like