Skip to main content

Roadmap

AutoAir is a visual and code-based DAG builder for Apache Airflow 3. It aims to cover what Airflow 3 can express, deploy the way production Airflow reads DAGs, and leave access control to Airflow. Released features are listed per version; planned items have no dates and may change. Detailed changes are in the changelog.

Released in 0.1.0​

Builder​

  • Visual editor inside the Airflow UI (Airflow 3.0 and later) and as a standalone server.
  • Task groups, notes, insert-on-edge, copy and paste with XCom remapping, undo and redo.
  • Command palette, context menus, keyboard shortcuts and automatic layout.
  • Run-state and version-diff overlays on the canvas.
  • Full-screen code mode and an editable Code tab that apply edited Python back to the DAG, with a read-only preview of the file each environment gets.
  • Raw code DAGs: hand-written DAG files stored, reviewed, validated and deployed like visual DAGs.
  • English and Turkish UI, light and dark themes, and configurable branding.

Authoring​

  • Typed DAG spec with automatic migrations as the single source of truth.
  • Operator catalog of every installed provider, plugin and package, read statically without importing modules.
  • TaskFlow decorators, dynamic task mapping, Airflow 3.4 task loops and setup/teardown tasks.
  • Callbacks and notifiers, retry policies, deadline alerts and human-in-the-loop operators.
  • Schedule editor for cron, presets, multi-cron, assets, asset watchers, partitions and combined schedules, with a next-runs preview.
  • DAG params with a trigger-form preview, and environment overlays for dev, staging and prod.
  • Jinja-aware editors with completion, drag-to-map XComs and typed connection forms.
  • Python and shell editors with completion, hover, signature help and live lint.
  • Output as Python, a JSON spec with a loader, Airflow 3.4 YAML or dag-factory YAML.
  • Import from Python DAG files, from Airflow, from YAML, and in bulk from a folder or a git repository.

Validation and policy​

  • Live validation of the graph, operator parameters, Jinja and the target Airflow version.
  • Import check of the generated DAG with a warm DagBag pool.
  • Organisation policy switches, declarative policy.rules and an optional Open Policy Agent decision.
  • Detection of hard-coded secrets in parameters and inline code.

Deploy and lifecycle​

  • Deploy to a local DAG folder, to object storage, or to Git through the git CLI, GitHub, GitLab, Bitbucket Cloud, Azure DevOps or Gitea.
  • Pull-request mode or direct push, with labels, reviewers, drafts and GitHub App authentication.
  • Deploy target discovery from Airflow's DAG bundle configuration, switchable without a restart.
  • Commit and pull-request templates with a required context block and git trailers.
  • Optimistic locking with a three-way semantic merge, version history and a visual graph diff.
  • Environment promotion, and per-environment connection and variable mapping.
  • Test runs of a task or a DAG without the metadata database; trigger, backfill, clear and pause.
  • Template gallery, reusable fragments, versioned operator templates with their own editor, and signed marketplaces.
  • Per-deploy git tokens typed by the user, never stored.

Collaboration, identity and access​

  • Real-time co-editing with presence, remote cursors and per-user undo.
  • Roles and permissions on top of Airflow's auth manager (FAB, Keycloak, SimpleAuthManager).
  • Four-eyes approvals for deploys and roles allowed to push directly.
  • Single sign-on through the Airflow session, Airflow credentials or an API token.
  • Audit trail with configurable sinks and CSV export, and an admin panel.

AI and MCP​

  • Generate, edit, explain, fix, review and document DAGs on pydantic-ai with Anthropic, OpenAI and compatible servers, Azure OpenAI, Gemini, Vertex AI, Ollama, Bedrock, Groq, Mistral or Cohere.
  • Troubleshooting, developer and admin chat assistants and retrieval; troubleshooting can run the Airflow actions it proposes after approval.
  • A documentation knowledge base over the Airflow, provider and Astronomer docs, usable without the AI extra and air-gapped.
  • Guardrails, an AI audit log, and an optional separate AI service.
  • MCP server for coding agents (stdio or HTTP), and an MCP client for external servers.

Operations​

  • Run modes: Airflow plugin, standalone server, and local mode on a workstation.
  • State in files, SQLite or PostgreSQL, with stateless replicas on PostgreSQL.
  • Configuration from environment variables, YAML, JSON, INI, Python files and airflow.cfg; feature flags switchable at runtime from the admin panel.
  • Health, readiness and Prometheus endpoints, JSON logs and optional OpenTelemetry tracing.
  • Container image, Helm values for the official Airflow chart, and a Docker Compose demo.
  • Supported versions: Apache Airflow 3.0 to 3.3 (3.4 pre-release tested), Python 3.10 to 3.14.

Released in 0.2.0​

Details and upgrade notes: changelog.

  • Editing existing Python DAGs. Source policies (generate, preserve, strict-preserve) patch a hand-written file with LibCST instead of regenerating it, and deploys detect files changed outside AutoAir. See Editing existing Python DAGs.
  • Astronomer Blueprint steps. See Blueprint steps.
  • Large graphs. Faster validation and a low-detail overview for DAGs with 500 to 1,000 tasks. See Large DAGs.
  • Approving agent tool calls. See Tool approval.
  • Isolated import checks. A separate validator service with no network access. See Validator service.
  • Airflow 3.4. DAG-page tab scoping, reparse and retry changes; tested against the 3.4 pre-release until 3.4.0 is final.
  • Plugin-mode visual baselines. Visual regression screenshots taken inside the Airflow UI.

Planned or under consideration​

Builder and editor​

  • Runtime frontend extensions. Load custom node kinds into the UI at runtime instead of registering them at build time.

Security and operations​

  • Managed Airflow guides. Deployment guides for managed Airflow offerings.

Testing​

  • PostgreSQL in the CI matrix. Run the Python and Airflow version matrix against the PostgreSQL state store. One PostgreSQL job covers it today.
  • Live provider checks. GitHub, Bitbucket Cloud, Azure DevOps and Gitea are covered by contract tests against fake servers, and the AI features by a scripted model client. GitLab, FAB and Keycloak are already checked against real servers; add real GitHub, Bitbucket Cloud, Azure DevOps, Gitea and LLM provider runs.

Proposing a feature​

  • Open a feature request describing the problem, who has it and what you do today instead.
  • Use GitHub Discussions for open-ended ideas and questions before they become a concrete request.
  • Check the open issues first, and add your use case to an existing request rather than opening a duplicate.
  • To contribute the change yourself, read Development and say in the issue that you are working on it.