Compatibility​

• Apache Airflow 3.0 or later, below 4 (CI: 3.0.6, 3.1.8, 3.2.2, 3.3.2 and the 3.4 pre-release). Generated imports follow the target Airflow version.
• Python 3.10 to 3.14.

Highlights​

Builder and canvas. A visual editor inside Airflow (Browse → DAG Builder and an AutoAir tab on each DAG page from Airflow 3.1; a full page at /autoair/ui/ on 3.0) or as a standalone server, with task groups, notes, a command palette, copy/paste with XCom remapping, auto-layout, run and diff overlays, full-screen code mode and English and Turkish UIs. See the user guide.

Authoring. Every installed operator through a static catalog; Python references, inline functions and embedded scripts; TaskFlow decorators, dynamic mapping, task loops (3.4), setup/teardown, callbacks, retry policies (3.3), deadlines (3.1) and human-in-the-loop operators; DAG params with a trigger-form preview; typed connection forms and per-environment connection mapping. Output as Python, a JSON spec with a loader, Airflow 3.4 YAML or dag-factory YAML; import from Python files, from Airflow, from YAML and from whole repositories.

Scheduling. DAG time zones, cron, multi-cron, asset expressions, event-driven assets, assets-or-time, assets-and-time and partitioned schedules, with a live next-runs preview. See scheduling and assets.

Validation and policy. Live graph, catalog, Jinja, policy and version checks plus a warm DagBag pool; organisation policy switches, declarative policy.rules and an optional Open Policy Agent decision. See policy.

Lifecycle and deploy. Local, object-storage and Git targets (git CLI, GitHub, GitLab, Bitbucket, Azure DevOps, Gitea), pull-request or push mode, a deploy-target dialog that discovers where Airflow reads DAGs, mandatory governance context and git trailers, three-way merge on conflicts, version history with a graph diff, test runs, and trigger, backfill, clear and pause from the editor. See deploy.

Collaboration and access. Real-time co-editing; IAM on top of Airflow's auth manager with built-in and custom roles, four-eyes approvals and direct-push roles; SSO through the Airflow session; an audit trail with configurable sinks; an admin panel. See IAM.

AI and MCP. Generate, edit, explain, fix, review, document and troubleshoot on pydantic-ai with Anthropic, OpenAI-compatible, Azure OpenAI, Gemini, Vertex AI, Ollama, Bedrock, Groq, Mistral or Cohere models, central or per user; agentic workflows, retrieval, a documentation knowledge base, guardrails and an AI audit log; a separate AI service; an MCP server and client. See AI.

Operations. Plugin, standalone and local modes; file, SQLite or PostgreSQL state with stateless replicas; health and readiness (degraded start with /api/ready returning 503 when the deploy target cannot be built), Prometheus metrics, JSON logs and OpenTelemetry; a container image, Helm values and a Docker Compose demo with profiles. See administration.

Not yet verified against real services​

Real LLM provider APIs and the GitHub, Bitbucket Cloud and Azure DevOps APIs are covered by tests against scripted or fake servers; see the roadmap.