Windmill
Also known as: Windmill Labs, windmill.dev, Windmill AI
Open source code first workflow orchestration platform turning Python, TypeScript, Go, SQL and Rust scripts into auto generated UIs, scheduled jobs, webhook APIs and DAG workflows, with approval gates for AI orchestration and full self hosting.
Windmill is an open source, code first workflow orchestration and internal developer platform built by Windmill Labs, founded in 2022 out of Y Combinator's S22 batch and backed by Gradient Ventures and Bessemer Venture Partners. It runs on a Rust based job engine that the company benchmarks at roughly thirteen times the throughput of Apache Airflow as a self hostable engine, and it reports more than 4,000 organisations and over 300 enterprise customers including Zoom, Panther Labs, CFA Institute and Photoroom.
Its distinguishing move is turning ordinary scripts into infrastructure. Code written in Python, TypeScript, Go, Bash, SQL, Rust, PHP or C# becomes an auto generated user interface, a scheduled job, a webhook triggered API endpoint or a step in a multi step DAG, from one platform. On top sits a low code app builder plus a full code React and Svelte frontend builder for internal dashboards, so the same script can serve an engineer, an operator and an API consumer without being rewritten.
For agents it offers AI and language model workflow orchestration with approval gates, which makes human sign off an execution primitive rather than something the application implements. Triggering is unusually broad for this category, covering cron schedules, webhooks, Slack, email, Kafka, Postgres change data capture and message queues. Operationally it provides secret management, granular user, group and folder role based access control, audit logs, detailed metrics and execution history, and Git sync so workflows are version controlled alongside everything else. It is SOC 2 Type II compliant and deploys either on the customer's own infrastructure or on Windmill Cloud.
One caveat is worth knowing before adopting the free build. An independent 2026 evaluation of seven open source workflow tools found that most, Windmill among them, are open core rather than fully open, with secrets management, user roles, audit logs and approver identity held back for paying customers. The reviewer's summary was that the paywall sits precisely on the security features, which matters most for exactly the regulated and security sensitive workloads this class of tool is often bought for.
Vendor details
Canonical URL
https://www.windmill.dev
Category
Agent infrastructure
Company status
independent
Use cases & customers
Target customers
Deployment options
In practice
A Python script needs to be a scheduled job, an internal tool with a UI and a webhook API all at once. Windmill generates the interface and the endpoint from the script's parameters rather than requiring three implementations.
An AI workflow must stop for human sign off before it acts. Approval gates are built into the orchestration layer rather than being something you implement in the application.
Compliance rules out a managed automation service. The platform self hosts on your own infrastructure with role based access control, secret management and audit logs, though check which of those sit behind the paid tier.
Sources & related URLs
Agentic Index coverage score
10.0 / 14 capabilities · 71%
| Integrations & Tool CallingExtensive integrations plus native connectivity to Kafka, Postgres change data capture, webhooks, email and message queues, GitHub sync, Slack, and the ability to call any API directly from scripts in eight languages. 2026-08-05 | Full |
|---|---|
| Workflow OrchestrationScripts compose into multi step DAG flows and data intensive pipelines on a Rust based job engine benchmarked at roughly thirteen times Apache Airflow throughput, with dedicated AI agent and language model workflow orchestration. 2026-08-05 | Full |
| Knowledge Grounding & RAGNo retrieval, vector store or knowledge grounding capability is provided. Windmill orchestrates AI and machine learning workflows but grounding belongs to whatever the script calls. 2026-08-05 | Unable to verify |
| Human Oversight & GuardrailsAI agent and language model workflow orchestration ships with APPROVAL GATES, making human sign off an execution primitive rather than an application concern, with approver identity tracked. Rare in this pocket. 2026-08-05 | Full |
| Security, Identity & GovernanceSOC 2 Type II compliant, with secret management, granular user, group and folder role based access control, audit logs and permission controls. CAVEAT: an independent 2026 evaluation found that in the free open source build several of these, including secrets management, user roles, audit logs and approver identity, are held back for paying customers. 2026-08-05 | Full |
| Observability & AuditabilityDetailed logs, metrics and audit trails, real time monitoring, and full execution history for debugging, presented as first class platform capabilities rather than an add on tier. 2026-08-05 | Full |
| Memory & State PersistenceA Postgres backed engine persists execution history and job state, with secret management and Git synced workflow versioning, but no agent scoped memory or durable resume primitive is documented in the way temporal and trigger-dev document theirs. 2026-08-05 | Partial |
| Deployment & Data ResidencyFully self hostable open source via Docker on any infrastructure, deployable to third party hosts, or run on Windmill Cloud, so the customer chooses where automations and sensitive data live. 2026-08-05 | Full |
| Prebuilt Agents, Templates & PacksReusable components let scripts be saved and shared across teams and auto generated UIs remove interface work, but no prebuilt agent catalogue or ready to deploy agent templates are documented. 2026-08-05 | Partial |
| Triggers & Channel CoverageUnusually broad trigger coverage for this pocket: cron schedules, webhooks, Slack, email, Kafka, Postgres change data capture and message queues, plus auto generated UIs and API endpoints as human and machine entry points. 2026-08-05 | Full |
| Model Flexibility & RoutingWindmill AI connects scripts to language models and AI APIs and is model agnostic by construction, but no model registry, routing policy, fallback chain or key management surface is provided by the platform. 2026-08-05 | Partial |
| APIs, SDKs & MCP ExtensibilityAn open source core with scripts in eight languages, webhook triggered API endpoints generated from any script, Git sync, a low code app builder and a full code React and Svelte frontend builder. No MCP server is documented. 2026-08-05 | Full |
| Testing, Debugging & OptimizationExecution history, detailed logs and metrics support debugging and Git sync gives reproducibility and rollback, but no evaluation harness, benchmark suite or agent quality regression testing is documented. 2026-08-05 | Partial |
| Browser & Computer UseNo browser control or computer use capability is documented. The platform executes scripts and API calls rather than operating interfaces. 2026-08-05 | Unable to verify |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
Windmill version 1.789.0 introduces an EXIT_AFTER_N_JOBS worker mode to manage environment cleanup by automatically restarting worker processes, alongside new memory limits for Go build subprocesses. The release also automates the process of building binaries to object storage upon deployment and adds the ability to open AI sessions directly from runs, jobs, and trigger pages.
Bears on: Workflow orchestration
View sourceWindmill released version 1.783.0, introducing public sharing options for job pages and git-sync pull request diff checks. The update also adds more development workspace environment labels and the ability to open session edits directly in the preview panel.
Bears on: Workflow orchestration
View sourceWindmill version 1.779.0 introduces raw application deployment directly from sources and custom worker tags for database jobs. The release also expands Model Context Protocol (MCP) support by exposing createApp and updateApp as full-code app tools and clarifying which MCP endpoint tools a token scope exposes.
Bears on: MCP / tool calling / API
View sourcePricing
Free and open source when self hosted; Windmill Cloud offers a free tier with paid tiers above it and custom enterprise terms
execution volume and seats, or free if self hosted
What is public
The open source self hosting route and the existence of free and paid cloud tiers are public; specific paid rates were not retrieved.
Billing mechanics
Two routes. The open source edition is free to self host on any infrastructure, where the only cost is compute and operations. Windmill Cloud starts with a free tier for small projects and scales through paid tiers adding team collaboration, priority support and higher execution limits, with enterprise terms negotiated.
Cost watchouts
The free open source build is open core: secrets management, user roles, audit logs and approver identity are reported as paid features, so the security capabilities are exactly what the paywall gates. Self hosting also carries real infrastructure cost, modelled by one host at roughly 5 to 15 dollars a month for small deployments before scaling.
Variable cost rationale
Self hosting caps vendor spend at zero and moves cost to infrastructure the team already operates, and even on cloud the metering is execution based rather than per seat, so exposure is the lowest in this pocket provided the required security features are available in the chosen tier.
Additional watchouts
An independent 2026 evaluation of seven open source workflow tools found the security features are precisely what most of them, Windmill included, hold back from the free build. Any team adopting the open source edition for a regulated or security sensitive workload should confirm secrets management, roles, audit logs and approver identity are actually available in the tier they intend to run.
Sales call required
Mixed (some tiers require a call)
Free / trial
Fully free open source self hosted edition, plus a free Windmill Cloud tier for small projects
Commercial notes
Positioned against Apache Airflow on performance, with a claimed thirteen times throughput advantage as a self hostable engine, and against managed automation platforms on control. For teams that already run infrastructure, the self hosted route makes the platform cost effectively zero, which is a materially different commercial shape from the per run and per compute second models elsewhere in this pocket.
Key ambiguities
No first party rate card was retrieved in this pass, so paid tier prices and enterprise terms are unknown, and the precise boundary of which security features are gated in the free build is reported by an independent reviewer rather than published by the vendor.
Missing data
Paid tier rates, enterprise pricing, execution limits per tier, and the vendor's own statement of which features are enterprise gated.
Related vendors
- Acrab — Singapore compute infrastructure company building a full stack…
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- AIsa — Unified resource and payment gateway for AI agents that lets them…
- AlphaBitCore — AI control plane that governs how models, agents, tools, and…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
Alternatives to Windmill
The closest documented capability profiles to Windmill among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Thread AI10.0 / 14Adds documented Knowledge Grounding & RAG
- Bernstein10.5 / 14Fuller documented coverage on Model Flexibility & Routing and Testing, Debugging & Optimization
- Inngest10.5 / 14Fuller documented coverage on Memory & State Persistence and Testing, Debugging & Optimization
- Temporal9.5 / 14Fuller documented coverage on Memory & State Persistence
- Tray.ai11.5 / 14Adds documented Knowledge Grounding & RAG
- Trigger.dev10.5 / 14Adds documented Browser & Computer Use
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded