[feature] Add native OpenAI Agents SDK support

[Muhammad-usman92]

What

Adrian currently supports LangChain/LangGraph agents through SDK-side instrumentation. This feature would add support for agents built directly with the OpenAI Agents SDK, including Agent, Runner.run, function_tool, handoffs, and guardrails.

Why

This would let OpenAI Agents SDK users send agent activity through Adrian’s existing monitoring pipeline without having to rebuild their agents on LangChain/LangGraph.

The initial goal would be trace-based monitoring support, mapping OpenAI Agents SDK execution into Adrian’s existing PairedEvent model and reusing the current backend, verdict, dashboard, and alerting flow.

Notes / context

Tagging @shahtajkhalid here since this came up as a potential contribution area.

##Proposed initial approach:

• Add adrian.adapters.openai_agents.
• Add an optional OpenAI Agents SDK dependency extra.
• Register an OpenAI Agents SDK tracing processor from adrian.init() when the Agents SDK is installed.
• Map generation spans to Adrian LLM events.
• Map function/tool spans to Adrian tool events.
• Map agent, handoff, and guardrail spans to context/metadata.
• Add tests for span-to-PairedEvent conversion.
• Add a small OpenAI Agents SDK example.

This first phase can focus on monitoring only and should not require major backend changes, since Adrian already has a generic PairedEvent ingest path.

A follow-up phase could explore runtime control support for block and hitl modes by adding pre-tool execution gating for OpenAI Agents SDK function tools, similar in goal to the existing LangGraph ToolNode control path.

I would be happy to contribute this if the direction aligns with the project roadmap.

[shahtajkhalid]

Hi @Muhammad-usman92, thanks for opening a feature request.

This feature is already in-progress, you can track it against: #40.

Thank you!