Introducing Coder Agents

Today marks the beta release of Coder Agents, a new way to run AI development workflows on the self-hosted infrastructure you already control. It gives platform teams centralized control over models, prompts, MCPs, skills, and network-isolated workspaces, enabling a first-class agent workflow experience for developers.

Use the Coder Agents UI or API to delegate tasks to agents running on your infrastructure

Coder provides the infrastructure behind how agents run and provision workspaces on demand, so teams can standardize workflows instead of pushing that responsibility onto individual developers. At its core, Coder Agents includes a native agent that runs on your control plane, with built-in control and observability over agent behavior, going beyond what third-party agents expose and giving teams a consistent way to define and manage how work gets done.

Most agent tools are tightly coupled to a single provider’s ecosystem, creating deeper lock-in as workflows grow. Coder Agents separates how agents run from which models they use, giving teams the flexibility to evolve without being tied to a single agent or LLM provider.

What’s in the beta release

Coder Agents is a conversational interface that developers use to collaborate with self-hosted coding agents

The beta release of Coder Agents includes the core capabilities teams need to standardize how agent workflows run across their organization.

• A conversational interface and API for turning ideas into executed code changes, whether in the foreground or as background tasks
  
• Centralized controls for models, prompts, and usage, so teams can define how agents behave across environments
  
• Extensible workflows through skills, MCP, and sub-agents, enabling agents to take on more complex, multi-step tasks

These capabilities work together as part of a single system. Models and prompts can be managed centrally, while workspaces are provisioned only when execution is required. Coder Agents runs as part of the Coder control plane, so teams can adopt it without introducing another service to manage.

For teams that want to go beyond the UI, the Coder Agents API can trigger workflows from CI/CD pipelines, GitHub Actions, Slack, and other systems.

For platform teams standardizing AI development

Most teams start using AI agents the same way: individual developers experiment with different tools, each tied to a specific model or provider.

That works early on, but it quickly leads to fragmentation. Teams end up managing multiple agent workflows, each coupled to a different ecosystem, just to give developers access to the latest models from providers like Anthropic, OpenAI, Google, Bedrock, or self-hosted endpoints.

Coder Agents is built for the point where teams need to move beyond that.

Coder Agents integrates with all major AI model providers and OpenAI compatible self-hosted LLM models

Instead of standardizing on a single tool or forcing developers into a single ecosystem, Coder provides infrastructure orchestration to standardize how agent workflows run while keeping model choice flexible. Intelligence continues to come from the models, but how agents execute, how workspaces and compute are provisioned, and how behavior is controlled become consistent across the organization.

This is especially important in environments where control matters, including regulated industries, air-gapped deployments, and organizations that cannot rely on SaaS-based tooling.

Two ways to run agents on Coder

Coder Agents is a native way to run agents on your infrastructure, but it is not the only way to use agents with Coder.

Teams can still run third-party agents like Claude Code or Codex inside Coder Workspaces. This is especially useful for organizations that have already built workflows around those tools. Running them on Coder provides a safer, more controlled environment today, while giving teams a glidepath to adopt a more centralized and flexible approach to running agent workflows via Coder Agents over time.

The difference is how those agents are deployed. Third-party agents run inside individual workspaces while Coder Agents runs through the Coder control plane with centralized control over models, prompts, and execution.

Because of that, teams running third-party agents on Coder often pair them with AI Governance to manage network access, enforce policies, and improve visibility into how those agents run inside workspaces.

Introductory access through September

Beginning today, Coder Agents is available in beta, with introductory access that includes full Coder Agents Premium features and no usage-based limits through September.

This introductory period is designed to get Coder Agents into the hands of more teams so they can build on the feedback that has already shaped the product. We want people using it in real workflows, sharing what is working well, and helping guide its continued evolution. Join us in Discord or reach out your customer success manager to let us know what’s working well for you and how we can help.

Coder Agents is already being used by early customers, but it is still maturing. Ahead of release, it underwent a security review, and over the coming releases, we will continue improving scalability, API flexibility, and overall performance. The Coder Agents API is functional in beta today but may evolve, including potential breaking changes, as we refine the platform. As always, we will communicate changes clearly and give teams time to adapt.

A note on Coder Tasks

Coder Agents is the long-term direction for orchestrating agents on Coder infrastructure, and over time, it will replace Coder Tasks. That change is not immediate. We will share the plan clearly, give customers and community members time to adapt, and provide a migration path before any deprecation takes effect. Before Tasks is removed, we will make sure customer workflows can transition to Agents with a better experience.

Get started

Coder Agents is available now in beta through our September release.

Install Coder to try it on infrastructure you control, read the docs for setup and architecture details, or talk to your customer success team if you have feedback or would like to take a deeper dive.