windmill-labs/windmill

Code-first platform for workflows, apps and AI agents, on the fastest self-hostable orchestrator (benchmarked against Airflow, Prefect and Temporal). Business teams build with AI, engineers control what ships. Open-source alternative to Lovable, Retool and n8n.

What is Windmill?

Windmill is an open-source platform you host yourself that lets you build workflows, web apps, and AI agents by writing code. It's described as a self-hostable alternative to tools like Retool, n8n, and a simplified Temporal.

Core idea: scripts become everything

You write a small, focused script (a plain function) in a language like Python, TypeScript, Go, or Bash. Windmill then automatically turns that script into:

  • A shareable UI (a web form generated from the script's parameters)
  • An API endpoint
  • A cron job (scheduled task)
  • A CLI command

You can chain scripts together into flows (workflows), build full-code React or Svelte apps on top, or compose them into AI agents and data pipelines.

How it works

  1. Write a script — in the browser-based editor or locally with the CLI/VS Code extension, synced with Git.
  2. Windmill generates the frontend — your script's typed arguments become a form, an HTTP endpoint, and a CLI command automatically.
  3. Chain scripts into flows — visually or as code, with each step running as its own job with its own logs.
  4. Build UIs — on top of your scripts and flows.

AI with guardrails

Business teams and engineers work on the same instance. AI sessions let non-engineers build scripts, flows, and apps in a workspace fork with a live preview and a diff to review before deploying. Protection rulesets can block direct deployment, so changes reach production only after an engineer reviews them. AI-built code runs on its own worker groups or a separate instance.

Data handling

  • Data tables — each workspace gets managed Postgres tables, queried with type-checked SQL without exposing connection strings.
  • DuckDB — scripts can query S3/Azure files (CSV, Parquet, JSON), data tables, Postgres, MySQL, and BigQuery in SQL.
  • DuckLake — stores large datasets as lakehouse tables in workspace object storage, queried in plain SQL.
  • Pipelines (alpha) — wire scripts by the data they read/write instead of hand-drawing a DAG.

Technical stack

  • Database: PostgreSQL (compatible with Aurora, Cloud SQL, Neon, Azure PostgreSQL)
  • Backend: Rust — stateless API servers and workers pulling jobs from a Postgres queue
  • Frontend: Svelte 5
  • Sandboxing: nsjail and PID namespace isolation
  • Runtimes: Bun (TypeScript/JavaScript), Python with uv, plus Go, Bash, PowerShell, Rust, C#, Java, Ansible, and more

Performance

Windmill claims to be the fastest self-hostable workflow engine, beating Airflow, Prefect, and Temporal in benchmarks (both 40 lightweight tasks and 10 long-running tasks). Once a job starts, there's no overhead vs. running the script directly; the added latency for queueing and result reporting is ~50ms.

Security

  • Sandboxing via nsjail for filesystem/resource isolation, plus PID namespace isolation to prevent jobs from accessing worker process memory.
  • Secrets: one encryption key per workspace for stored credentials.

Self-hosting

  • Docker compose — three files, one command: docker compose up -d
  • Kubernetes — Helm charts available
  • Cloud providers — works on AWS, GCP, Azure, Fly.io, Render.com, Hetzner, Digital Ocean, etc. Rule of thumb: 1 worker per 1vCPU and 1–2 GB RAM.
  • OAuth/SSO — configure Google Workspace, Microsoft/Azure, Okta from the superadmin UI.

License

  • Community Edition is free to use internally or via APIs, within set limits.
  • Selling/reselling it as a managed service, modifying/wrapping it, or re-exposing it as a feature of your product requires a commercial license.
  • The binary built without the enterprise flag is AGPLv3; API clients and OpenAPI spec are Apache 2.0.

Contributing

The project is not seeking outside contribution at this time, though bug reports, feature requests, and small trivially-verified PRs are welcome.

Key facts

  • 20+ languages supported for scripts
  • Scripts can be triggered by schedules, webhooks, HTTP routes, Kafka, WebSockets, emails, and more
  • Existing dbt projects run unmodified as scripts
  • Default Docker credentials: admin@windmill.dev / changeme
  • Active project: high commit activity, Discord community, and a Hub for sharing community scripts

In short: write a function, get a UI, an API, a schedule, and a workflow — all self-hosted, with AI assistance and human review built in.

Related

  • Project
    emcie-co/parlant

    An interaction control harness for customer-facing AI agents that uses dynamic context engineering to ensure consistent, compliant, and brand-aligned behavioral control.

  • Project
    langchain-ai/langchainjs

    A framework for building LLM-powered applications that simplifies the integration of models, data sources, and components through a modular, interoperable architecture.

  • Project
    langbot-app/LangBot

    Open-source platform that unifies LLMs (OpenAI, DeepSeek, local Ollama, etc.) into AI agents deployable across Slack, Discord, Telegram, WeChat, QQ, and more—with a web dashboard, plugin marketplace, and MCP/CLI tools so human or AI developers can manage every bot from one place.

  • Project
    gastownhall/gastown

    Gas Town is an open‑source Go‑based workspace manager that lets you run many AI coding agents (Claude, Copilot, Codex, Gemini, etc.) on multiple Git projects while persisting every step in a Git‑backed ledger. A central AI coordinator (the *Mayor*) creates *convoys* of work items (*beads*), hands them to worker agents (*Polecats*), and a set of watchdog services (Witness, Deacon, Dogs) keep the system healthy. Completed work is merged via a Bors‑style queue (*Refinery*), and an optional federation layer (*Wasteland*) lets different installations share tasks. Installable via Homebrew, `go install`, or Docker, it provides a CLI (`gt`) for creating rigs, crews, convoys, and for monitoring progress.