Terraform your dev environment, then hand the keys to an AI

What it does

Coder is a self-hosted control plane that provisions developer workspaces and AI coding agents through Terraform templates. It connects users to remote environments over Wireguard tunnels and supports standard IDEs like VS Code and JetBrains. Workspaces can live on EC2, Kubernetes, Docker, or anything else Terraform can touch, and the platform automatically shuts them down when idle to cut costs. It also runs AI agent loops in the control plane so LLM credentials never live inside the workspace itself.

The interesting bit

Most remote development tools stop at the IDE; Coder also tries to be the infrastructure layer for automated agents. The AI loop runs in the control plane, attaching user identity to every action and keeping API keys out of the workspace entirely. It is a bet that your infrastructure for human developers and their automated counterparts should be the same thing.

Key highlights

• Workspaces are pure Terraform, so you define EC2 VMs, Kubernetes Pods, or Docker containers as code.
• Native AI agent support with bring-your-own-model flexibility, including self-hosted LLMs.
• Agent loops execute in the control plane, keeping LLM credentials, governance, and audit logs centralized.
• Automatic idle shutdown to save infrastructure costs.
• Wireguard-based tunneling and plugins for VS Code, JetBrains Toolbox, and Gateway.

Caveats

• Production deployments require PostgreSQL 13+ and an external access URL; the built-in quickstart mode is not production-grade.
• Some capabilities are reserved for the Premium plan.
• The README does not detail AI agent maturity or workspace performance limits.

Verdict

Platform teams already managing Terraform will find Coder a natural fit for offering remote development and AI agents without surrendering data to a SaaS. Solo developers just looking for a quick cloud IDE will likely find the infrastructure overhead excessive.