Temporal AI Framework Integrations Are GA: What the Announcement Says and What the Registries Showed the Same Day

Temporal announced on October 30 that its integrations for several AI agent frameworks are generally available. The idea is sound: run the agent loop inside a Temporal Workflow so a crashed Worker, a failed model call or a long wait for a person does not lose progress. But on the day of the post, the public package metadata I checked did not uniformly say "GA". If you plan to adopt one of these, check the exact package you will install.

This is a read of the announcement, the integrations repo README and the PyPI and npm registries. I did not install or run any integration.

What the announcement says

The post, by Hien Phan, Johann Schleier-Smith and David Hyde, lists the integrations it calls generally available:

Integration	Languages
Google ADK	Python, Go, TypeScript
OpenAI Agents SDK	Python (already GA), TypeScript
AWS Strands	Python, TypeScript
LangGraph	Python
LangSmith	Python, TypeScript
AI SDK by Vercel	TypeScript
Gemini SDK	Python

Its description of how they work:

• Workflows hold long-lived coordination logic. Activities do model calls, API requests, database access and other side effects.
• State that crosses that boundary is serialized so Temporal can record it. After a Worker restart or deployment, Temporal rebuilds the Workflow from its Event History, and recorded Activity results are reused instead of repeating the model or tool call.
• The integrations turn off, or ask you to turn off, framework-level retry loops so Temporal manages retries and timeouts in one place. They keep clients and credentials on the Worker and carry trace context across processes.
• Where supported, they can stream tokens through Workflow Streams, which the post says is in Public Preview for Python and TypeScript. Agents can wait for human approval without polling or holding a Worker open.
• All are built on Temporal's Plugin API. The post names partner-built integrations for Pydantic AI, Mastra and Braintrust.

The post's Python example for Google ADK swaps the model name for TemporalModel("gemini-2.5-flash"), which runs each model call as an Activity, and registers GoogleAdkPlugin with the Client and Worker. It names OpenAIAgentsPlugin for the OpenAI Agents SDK, TemporalAgent for Strands, and execute_in labels for LangGraph nodes.

It also says that if you are upgrading, you should review the release notes before updating running Workflows, since there are package and import changes, breaking changes, and cases where Worker Versioning may be needed.

What I checked the same day

I checked the following on October 30, 2026, around 21:24 IST. Registries and READMEs change, so treat this as a snapshot, not a verdict.

The repo README. The temporalio/ai-integrations README lists a status per plugin. As I read it, only the OpenAI Agents plugins (Python and TypeScript) and the Vercel AI SDK TypeScript plugin carry no preview label. Most others are marked "Public Preview" or "Pre-release", including python/google_adk (Pre-release), python/langgraph, python/langsmith, python/google_genai, python/strands_agents, go/googleadk, typescript/google-adk and typescript/langsmith (Public Preview), and typescript/strands-agents (Pre-release).

PyPI classifiers and versions.

• temporalio-openai-agents 1.1.0: "Production/Stable".
• temporalio-google-adk 0.0.1: "Alpha", the only release.
• temporalio-langgraph 0.1.0: "Beta".
• temporalio-strands-agents 0.1.0: "Beta".
• temporalio-langsmith 0.1.1: "Beta".
• temporalio-google-genai 0.1.0: "Beta".

npm.

• @temporalio/openai-agents latest 1.24.0.
• @temporalio/vercel-ai-sdk latest 1.0.0 (2 versions), published October 17.
• @temporalio/google-adk latest 0.1.0 (2 versions).
• @temporalio/strands-agents latest 1.24.0, @temporalio/langsmith latest 1.24.0.
• A separate @temporalio/ai-sdk package also exists, latest 1.25.0 with 29 versions, modified October 27. I did not work out how it relates to @temporalio/vercel-ai-sdk.

How to read the gap

A launch post and a README can be updated at different times. The README or classifiers may simply not have caught up with the announcement, and I cannot tell which is the lagging source. What I can say is that on the same day, the metadata I could read described the Google ADK Python package as a 0.0.1 alpha, and the post calls the Google ADK integration generally available. Also, a version of 0.x or a Beta classifier does not by itself say what Temporal's support terms are.

So before adopting, read three things for the specific package: its own release notes, its version and classifier, and Temporal's support terms for it. Pinning an exact version matters here, because the post itself warns of breaking changes between versions.

What I would check

These are my untested suggestions.

1. Pin the exact package version and read its release notes before upgrading running Workflows.

2. Run a failure drill. Kill a Worker mid-agent-run and see whether the run resumes without repeating a model call that was already recorded.

3. Look at retries. Confirm the framework's own retry loop is off, as the post says it should be, so two layers do not retry the same failing call.

4. Review what goes into Event History. Serialized agent state is recorded, so check what data (prompts, tool results) ends up there before you use real customer data.

5. Test a human approval wait that lasts longer than your deployment cycle.

6. Check the Workflow Streams preview separately if you need token streaming.

What I would not conclude

• That every listed integration is production-ready because the post says GA. On the same day, the README and PyPI metadata I read were not uniform.
• That the metadata is right and the post is wrong. It may just be lagging.
• That durable execution makes agent actions safe to repeat. The post says recorded results are reused on replay. I did not test what happens with a side effect that happened but was not recorded.
• That the code snippets work as shown. I did not run them, and I did not verify the import paths.

Frequently asked questions

What did Temporal announce?

Per its October 30 post, general availability of integrations for Google ADK, OpenAI Agents SDK, AWS Strands, LangGraph, LangSmith, Vercel AI SDK and Gemini SDK.

How does durable execution help an agent?

Per the post, the agent runs in a Workflow, side effects run as Activities, and after a failure Temporal rebuilds the Workflow from Event History and reuses recorded Activity results.

Where do the integrations live?

In the temporalio/ai-integrations repository on GitHub, with a separate package for each integration and language, per the post.

Which package is clearly stable?

On the day I checked, temporalio-openai-agents was classified Production/Stable on PyPI. Others I checked were Alpha or Beta.

Do I need to rewrite my agent?

The post says the integrations work with existing agent code, with framework-specific imports or annotations. I did not test that.

Is token streaming GA?

The post says Workflow Streams is in Public Preview for Python and TypeScript.

What should I read before upgrading?

The release notes. The post says they list package and import changes, breaking changes, and cases where Worker Versioning may be needed.

Sources

All read on October 30, 2026.

• Temporal AI framework integrations are now GA (Temporal, October 30, 2026)
• temporalio/ai-integrations and its README
• PyPI JSON metadata for temporalio-openai-agents, temporalio-google-adk, temporalio-langgraph, temporalio-strands-agents, temporalio-langsmith and temporalio-google-genai
• npm registry entries for @temporalio/openai-agents, @temporalio/vercel-ai-sdk, @temporalio/google-adk and @temporalio/ai-sdk

Not read: the release notes, the AI Cookbook, the Plugin author guide, and the source code of any integration.