Coder — Self-Hosted Cloud Development Environments and AI Agents
Define dev environments in Terraform, connect via secure tunnel, auto-shutdown idle resources, and run AI coding agents on your infrastructure.
- Source repo
- coder/coder
- Stars
- ★ 17k
- Last updated
- today
- License
- AGPL-3.0
- Primary language
- Go
- FA score
- 62/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- Developers need a consistent, reproducible dev environment defined in Terraform, launched in seconds via Docker templates.
- Not a fit if
- Small teams unwilling to run PostgreSQL and Terraform infrastructure
- Teams that need a fully offline setup with no external model providers
- Individual developers wanting a quick Windows-only trial
- Source review
- 62/100 · Some gaps
What does this agent do, and when should you use it?
Coder is a self-hosted platform for cloud development environments and AI coding agents. Workspaces are defined with Terraform and can be provisioned on EC2 VMs, Kubernetes Pods, Docker containers, etc. It connects through a secure Wireguard® tunnel and automatically shuts down idle resources to save costs. Coder Agents run a native AI coding agent whose loop executes in the control plane on your infrastructure, with no API keys in workspaces. It brings any model (Anthropic, OpenAI, Google, Bedrock, self-hosted) and offers centralized model governance, cost tracking, and audit logging. The platform includes a CLI, web UI, VS Code extension, and JetBrains plugin.
Coder runs with the coder server command, which starts a web server and CLI. It uses Terraform templates to define infrastructure (e.g., EC2, Kubernetes, Docker) for workspaces. It automatically shuts down idle resources. Coder Agents execute AI coding agent loops in the control plane, supporting any model provider, and workspaces never contain API keys. The AI Gateway centralizes authentication, auditing, and cost controls. It integrates with VS Code, JetBrains Toolbox, Dev Containers, and Kubernetes log streaming, and offers a GitHub Action for setup.
- Developers need a consistent, reproducible dev environment defined in Terraform, launched in seconds via Docker templates.
- Teams want to run AI coding agents like Claude Code on their own infrastructure to keep code within their network.
- Admins need centralized governance of AI model access, cost tracking, and audit logs.
- Organizations want to reduce cloud costs by automatically shutting down idle workspaces.
- Developers using JetBrains IDEs can open workspaces via the JetBrains Toolbox plugin.
How do you install or deploy this agent?
Install via script on Linux/macOS: curl -L https://coder.com/install.sh | sh. For Windows, download the binary or installer from GitHub Releases. Start the server with coder server and open http://localhost:3000.
How do you use this agent?
Run coder server to launch, then create your initial user via the web UI. For production, add PostgreSQL (13+) and an external access URL: coder server --postgres-url <url> --access-url <url>. Use coder --help for all flags and environment variables.
What are this agent's strengths and limitations?
- Self-hosted, keeping all data and workspaces within your infrastructure.
- Workspaces are defined in Terraform, making them portable across EC2, Kubernetes, Docker, etc.
- Auto-shutdown idle resources to save on cloud costs.
- Model-agnostic AI agents (Anthropic, OpenAI, Google, etc.) without storing API keys in workspaces.
- Centralized AI Gateway for authentication, cost tracking, and audit logging.
- Requires Terraform and PostgreSQL, adding deployment complexity.
- AI agent features require additional configuration for model provisioning and governance.
- Wireguard tunnel setup may require network configuration.
- Windows support is via separate installer; not covered by the install script.
How does this agent compare with similar options?
GitHub CodespacesKey facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Coder — Self-Hosted Cloud Development Environments and AI Agents This agent | 62 · Some gaps | Self-hosted serviceFreemium | ★ 17k | today | Go | OpenAI API · Claude API |
| Warren | 76 · Good | Self-hosted serviceFree + model costs | ★ 503 | 2d ago | TypeScript | Claude Code |
| HarnessRouter Community Edition | 67 · Some gaps | Self-hosted serviceFree + model costs | ★ 3.1k | today | Python | Codex · Claude Code |
| OtoDock | 77 · Good | Self-hosted serviceFreemium | ★ 191 | 5d ago | Python | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README emphasizes secure tunnels, no API keys in workspaces, centralized model governance, and audit logging, indicating attention to least privilege and sensitive data handling. Workflows use minimal permissions (e.g., permissions: {} in aigateway-prices-refresh) and include user confirmation (e.g., price refresh PRs are never auto-merged). Data flow transparency is evident in documentation and the price refresh workflow. Dependency security is addressed via go.mod replaces and overrides (e.g., security overrides in package.json). External effects are managed via Slack notifications and PR creation in workflows. Rollback support is weak, not explicitly mentioned. Source attribution is limited due to unverified publisher. Deductions: insufficient evidence for rollback and source attribution.
Evidence shows: README and documentation structure are consistent, workflows have explicit failure handling (e.g., Slack notification on price refresh failure). Dependency availability is addressed via go.mod replaces and overrides, but not verified. Failure messages are explicit in workflows. Deductions: dependency availability not verified, failure messages not fully checked in code.
Evidence shows: README clearly identifies target audience (developers and AI agents), scenarios include cloud development environments and AI coding agents. Capability boundaries are evident via documentation (e.g., workspaces, templates, agents). Trigger precision is explicit in workflows (e.g., cron schedules). Environment fit is addressed via install scripts and Helm charts. Deductions: capability boundaries and trigger precision not deeply verified in code.
Evidence shows: Information architecture is clear (README with doc links), install notes exist (quickstart), naming is stable (module and package names). Examples and FAQ are provided via doc links. Known limitations are not explicitly listed. License is AGPL-3.0. Versioning and changelog are not explicit in provided files. Maintenance responsibility is evident via contribution guide and issue tracking. Deductions: insufficient evidence for known limitations, versioning, and changelog.
Evidence shows: Output usability is evident via documentation and CLI help. Marginal value is evident via AI agent and cost-saving features. Cost-benefit is evident via auto-shutdown of idle resources. Deductions: no specific cost data or user feedback provided.
Evidence shows: Claims are explicit in README but not independently verified. Cross-source corroboration is evident via OpenSSF badges and doc links. Fact-inference separation is not explicit in documentation. Deductions: limited claim traceability, fact-inference separation not explicit.
- Publisher identity is unverified; treat source attribution with caution.
- Rollback mechanisms are not explicit; further confirmation needed.
- Dependency security relies on replaces and overrides, but their security is not verified.
- Known limitations are not listed, which may affect deployment decisions.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Changelog Bug fixes - Prevent coderd crashes from unsupported Tailnet RPCs (#28429) (#28957, 82fe9d17) - Pin build-dylib to macOS 15 runner images on release/2.29 (#30496, 44ced694) - Server: Reject workspace proxy hostname prefixes (#27544) (#27614, 16ad5d9f) - Server: Validate GCP instance token audience (backport to 2.29) (#30190, 1563fab3) Documentation - Correct broken CLI commands and flags from drift sweep (backport to release/2.29) (#28176, 625d2566) - Fix slack webhook backtick (backport of #28085 to release/2.29) (#28177, 97980f9e) …View on GitHub ↗ -
Release notes excerpt
Changelog > [!NOTE] > This is a mainline Coder release. We advise enterprise customers without a staging environment to install our latest stable release while we refine this version. Learn more about our Release Schedule. BREAKING CHANGES - Tasks fully deprecated and removed > Tasks are fully deprecated and removed in Coder v2.38. > > Migration: Users who rely on Tasks can stay on the [v2.34 Extended Support Release](https://coder.com/changelog/coder-2-34) during the support period. - Experimental chat API mounts removed (#29548) …View on GitHub ↗ -
Release notes excerpt
> ## Stable (since October 14, 2026) Changelog Bug fixes - Server: Validate GCP instance token audience (#29765, 1d246291a8) Documentation - Documentation: Document RPM install on RHEL-family Linux (#29295, 944aa0e525) - Flatten lone index pages to single files (#30129, fb14cd85f5) (@nickvigilante) Compare: [v2.37.3...v2.37.4](https://github.com/coder/coder/compare/v2.37.3...v2.37.4) Container image - docker pull ghcr.io/coder/coder:2.37.4 Install/upgrade …View on GitHub ↗