Securely manage cloud development environments and AI agents within your own infrastructure
Configuring virtual machines or containers to start only when needed and shut down automatically when idle prevents unnecessary costs. Coder is a self-hosted platform that defines these cloud development environments with Terraform and connects them securely via Wireguard tunnels. It recently added the ability to run AI coding agents on your own infrastructure, allowing you to delegate coding tasks without exposing external API keys to workspaces.
Defining workspaces with Terraform
Workspaces are written as Terraform templates and can include various infrastructure such as EC2 VMs, Kubernetes Pods, and Docker containers. These templates bundle the IDE, dependencies, and configuration information required for development. When developers create a workspace based on this template, they can complete an onboarding process that previously took days in just a few minutes. Unused resources are terminated automatically, contributing to cost savings.
AI agent execution environment
Coder Agents provides the functionality to run AI coding agents on self-hosted infrastructure. The agent's execution loop is handled by the control plane, and no LLM credentials are stored inside the workspace. You can use various AI models, including Anthropic, OpenAI, Google, Bedrock, and self-hosted models. User identity is recorded for all tasks, and model governance, cost tracking, and audit logs are managed centrally.
Installation and operation
On Linux and macOS, you can install it using an installation script, while on Windows, you can download the binary from GitHub Releases. Running the coder server command starts a local server, allowing you to create an initial user in the browser, create a Docker template, and provision your first workspace. For production deployment, you must specify PostgreSQL 13 or higher and an external access URL. If this flag is not specified, a built-in database and an evaluation access URL are configured automatically.
Extensible integration ecosystem
You can open workspaces in existing editors using VS Code and JetBrains Toolbox plugins. You can build development environments on Docker, Kubernetes, and OpenShift using Dev Containers, and you can also configure the Coder CLI in workflows via GitHub Actions. In restricted network or air-gapped environments, you can utilize a self-hosted VS Code extension marketplace. It is also possible to extend workspace configurations through community templates and modules.
Considerations before adoption
Since this platform must be operated in a self-hosted environment, basic knowledge of infrastructure management is required. For production use, setting up a PostgreSQL database and an external access URL is essential, and you should determine the infrastructure scale by referring to validated architectures. If you use the AI agent feature, per-model cost tracking and audit logs are important, so you should thoroughly review the central management features. Coder Premium is a paid feature for large teams that includes dedicated support.
