Windmill Labs
Windmill

Build AI agents with full control and observability

For platform teams who deploy AI agents to production and need full control over permissions, monitoring and audit logs.

Build agent with any LLM provider including self-hosted models
Give your agents tools via MCP or custom scripts in 20+ languages
Run full Claude Code, Codex or Opencode in sandboxes at their full power
Track every decision, tool call and code execution with full observability

Trusted by 4,000+ organizations, including 300+ EE customers at scale:

ZoomZoomKahootInvesting.comCFA InstituteAxiansAxiansPhotoroomPavePanther LabsNocd

What is an AI agent on Windmill?

Windmill provides enterprise-grade infrastructure to build AI agents around any LLM, including self-hosted models. Give your agents access to tools written in any language, connect them to your databases and APIs, compose them into flows alongside scripts and workflows, or run them in isolated sandboxes.

Your agents run on the same platform as the rest of your infrastructure, with full observability on every tool call and LLM request, RBAC, audit trails and versioning built in.

Read the docs
AI agent as a step in a Windmill flow

Two ways to build AI agents

Use the visual AI builder with tools for structured agent loops, or run full Claude Code, Codex or Opencode inside sandboxes for maximum autonomy. Both can be used as steps of a flow.

Visual AI builder with tools

The AI agent step is a ready-to-use node in the flow editor. Build agent workflows visually by combining agent steps with branching, approval steps, error handling and any other flow step.

Learn more about the flow editor

Claude Code, Codex and Opencode in sandboxes

Run coding agents inside isolated sandboxes with full capabilities: bash, code writing, file editing. Cached snapshots for instant startup, persistent volumes across runs and scoped permissions.

Learn more about AI sandboxes

Give your agents the right tools

Agents can use Windmill scripts and MCP servers as tools. Scripts are written in TypeScript, Python, Go, SQL or any supported language. Windmill reads each script's typed function signature and generates the tool schema the LLM sees. MCP servers expose their own tool definitions and are auto-discovered when connected.

Learn more about agent tools

See it in action

An AI chatbot built entirely on Windmill. The frontend is a full-code React app, the backend is a flow that calls an AI model with Windmill scripts as tools.

Built for production

Every agent runs within a platform built for enterprise requirements: immutable versioning, role-based access control, full audit trails, SSO and self-hosting. No extra setup needed.

Versioning and deploy

Every agent version is immutable and addressable by hash. Deploy from the UI, the CLI or Git sync. Roll back to any previous version instantly. Sync your workspace with GitHub or GitLab and use your existing code review workflows.

Learn more about deployment and versioning

Full observability

Every run is logged with inputs, outputs, duration and status. Real-time execution logs stream as your agent runs. Set up error handlers to send alerts on failure. Audit trails track who ran what and when.

Learn more about observability

RBAC and permissions

Control who can view, edit and run each agent with role-based access control. Organize agents into folders with group-level permissions. Authenticate users with SSO. Every action is recorded in the audit log.

Learn more about RBAC

Frequently asked questions

Build your internal platform on Windmill

Scripts, flows, apps, and infrastructure in one place.