OpenAI + @Temporalio : Building Durable, Production Ready Agents
summary
Shows a working integration that makes OpenAI Agents SDK agents durable by running them inside Temporal workflows: agents call the OpenAI Responses API, list tools implemented as Temporal activities (e.g., get weather, get IP→location), and use an activity_as_tool helper to generate tool JSON. A worker crash and restart resumes the agent via event sourcing so LLM calls are not reburned (demo picks up after a killed process); dynamic tool dispatch uses dynamic activity and retries/LLM backoff are handled by Temporal. Latency to server is tens of milliseconds.
AI-written summary · about
chapters (12)
- 0:00OpenAI agents SDK & Temporal intro
- 1:19Why combine OpenAI & Temporal
- 3:26OpenAI agents SDK deep dive
- 7:39Temporal for AI apps
- 18:15Failures & retries in Temporal
- 20:01Demo agentic loop
- 35:53Dynamic activities & tools
- 45:37Integrating OpenAI SDK & Temporal
- 51:42Demo durable agents
- 1:00:14Orchestrating microagents
- 1:06:12Resources & future
- 1:09:45Q&A
related talks
14mTemporal, OpenAI, and the Future of Agentic AI
56mOpening Keynote
38mHow to Leverage Temporal for Agentic Orchestration and Workflows
26mBeyond Sandboxes: Architecting Durable Runtimes for AI Agents
31mTemporal Substrate Architecture: Building Persistent Identity for Autonomous Agents
26mTemporal: A Better Way to Build Modern Applications
40mAI Agents in Practice
28mDurable Code Execution With Temporal