kaxil opened a new pull request, #74723: URL: https://github.com/apache/airflow/pull/74723
Someone who has used Airflow but never Common AI had no single place to see what the provider can do. `features.rst` listed eight call-level settings. Operators, toolsets, sandboxing, retry policies, provider fallback, batch, durable execution, usage limits, other agent frameworks and tracing were spread across the sidebar, and nothing said which extra or Airflow version each one needs, or whether it is stable. This turns `features.rst` into a feature map: one row per feature, grouped by what the reader is trying to do. Each row gives a one-line description, a link to the page that documents it, the extra it needs, the oldest Airflow version it runs on, and whether it is stable. The page keeps its path and its `howto/features` label, so existing links still work. The landing page links it next to Why Common AI, and it moves to the top of the Guides group in the sidebar as "Feature map". Three linked sections had no label to link to, so this adds labels for the AI Model tab, the confidence gate and "Any other framework". Where each column comes from: - **Extra**: `[project.optional-dependencies]` in `pyproject.toml`, and the optional imports behind each feature (for example, `SQLToolset` and `@task.llm_sql` import `sqlglot` from the `sql` extra; the managed-agent toolset comes from the Amazon or Google provider's `common.ai` extra). - **Airflow**: the `AIRFLOW_V_3_x_PLUS` gates in the provider source (3.1 for approval gates and HITL review, 3.3 for tool approval and the cross-retry usage total, 3.4 for the AI Model tab), plus the Airflow 3.0 floors of the git and Modal providers behind the `skills` and `modal` extras, and the 3.3 floor `retry_policies.rst` states for retry policies. - **Stability**: `stability.rst` on current main, including today's graduations (#74679, #74686, #74704, #74707). A parameter of a stable operator counts as stable, because that page promises a stable feature's "public parameters, their defaults and the behaviour described below". So `usage_limits`, `capabilities=` and `cache_prompt` are stable, and the `shields` extra and code mode stay experimental. The AI Model tab is a plugin and is not named in either table, so it is experimental under the page's own rule. Tool-call logging has no row yet. It is on by default with nothing to configure, and #74694 is making `LoggingToolset` private and rewriting that page. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
