agents

Temporal makes its LangGraph, Google ADK, Strands and OpenAI Agents SDK integrations GA

Temporal announced on October 30 that its integrations for seven AI libraries and agent frameworks are generally available, so agents built with them can resume after crashes without repeating model or tool calls.

Written by Mara at Sig · Nov 1, 2026

Temporal said on October 30 that integrations for several agent frameworks are now generally available: Google ADK (Python, Go, TypeScript), the OpenAI Agents SDK (TypeScript joins Python, which was already GA), AWS Strands (Python, TypeScript), LangGraph (Python), LangSmith (Python, TypeScript), the Vercel AI SDK (TypeScript) and the Gemini SDK (Python).

The idea is that you keep your agent code and add Temporal around it. In the ADK example, the only change inside the agent is TemporalModel in place of the model name. Each model call then runs as a Temporal Activity. If a Worker restarts or a deploy interrupts the run, the Workflow replays from its history and reuses recorded results instead of calling the model or tool again.

The integrations also turn off (or ask you to turn off) the framework's own retry loops, so retries and timeouts live in one place. Token streaming through Workflow Streams is still in public preview. The packages moved to the temporalio/ai-integrations repo, and Temporal says to read the release notes before upgrading running Workflows, since some have import changes and breaking changes.

Why it matters

An agent that runs for an hour and dies at minute 50 usually starts over and pays for every call again. This gives you resume-from-where-it-stopped without rewriting the agent, and lets it wait days for a person's approval without a process sitting open.

Try this

uv add temporalio-openai-agents (or npm install @temporalio/openai-agents), run one of Temporal's AI Cookbook examples against a local dev server, kill the Worker halfway through, restart it, and check in the Event History that finished model calls weren't sent again.