This is a great addition, and I agree that it should be a separate provider.
+1 from me. I would also like to join Kaxil in co-sponsoring this provider. Regards, Pavan On Fri, 26 Jun 2026 at 21:33, Jarek Potiuk <[email protected]> wrote: > Looks great. We should like have similar features for OpenAI and others in > a similar fashion. It's good to have them as a standalone "independent" > feature from common.ai. > > Side comment: I think it might also be a good point to discuss how the > Sandbox proposal fits in. This proposal clearly shows that we need a > Sandbox **component** (I'm not even sure if it should be a provider or an > executor) to map the sandboxing use cases that come with the various LLM > cases, operators, or hooks we have in Airflow. Ideally we should have a > "sandboxing" proposal that covers all executors we have. > > J. > > > On Fri, Jun 26, 2026 at 4:12 PM Shahar Epstein <[email protected]> wrote: > > > +1, great initiative! > > > > > > Shahar > > > > On Fri, Jun 26, 2026 at 3:13 AM Kaxil Naik <[email protected]> wrote: > > > > > Hi Airflow community, > > > > > > I'd like to propose adding a new community provider for Anthropic's > > Claude > > > API: apache-airflow-providers-anthropic. > > > > > > Integration overview > > > --------------------------- > > > Claude is a widely used family of LLMs accessed through the official > > > "anthropic" Python SDK. The provider lets Dag authors call Claude > through > > > Airflow connections and operators rather than embedding SDK calls and > API > > > keys in task code. It includes: > > > > > > - A hook that builds the right client for the configured platform (the > > > first-party Anthropic API, Amazon Bedrock, Google Vertex AI, Claude > > > Platform on AWS, and Microsoft Foundry), including keyless Workload > > > Identity Federation. > > > - A deferrable Message Batches operator, sensor, and trigger for the > > > asynchronous, lower-cost batch workflow (submit, defer, collect). > > > - A deferrable Managed Agents session operator and trigger for > > Anthropic's > > > server-run agent sessions. > > > > > > It targets Airflow 3+ and would ship as incubation. It is > intentionally a > > > separate > > > provider rather than part of common.ai: common.ai is a > provider-agnostic > > > abstraction, whereas this exposes Anthropic-specific surfaces (Message > > > Batches, Managed Agents, token counting, the platform clients) that do > > not > > > fit a neutral interface. The two are complementary. > > > > > > > > > Working implementation > > > --------------------------------- > > > https://github.com/apache/airflow/pull/69003 > > > > > > > > > I am happy to steward and > > > sponsor; I know Pavan would too, and if another committer would like to > > > co-steward this provider, please let me know. > > > > > > Thanks, > > > Kaxil > > > > > >
