Coder Agents Launches in Beta: Self-Hosted AI Coding for Enterprises
Coder Technologies released Coder Agents in beta on May 6 — a self-hosted, model-agnostic AI coding agent platform for enterprises that need to keep code and prompts inside their own infrastructure.
Coder Technologies launched Coder Agents in beta on May 6, targeting enterprise engineering teams that want AI coding agents but can’t send source code or prompts to a third-party cloud service.
The pitch is straightforward: run the entire agent stack — control plane, orchestration, execution — on your own infrastructure. No code leaves the network boundary. The agent supports Anthropic, OpenAI, Google, AWS Bedrock, and any self-hosted OpenAI-compatible endpoint, so teams aren’t locked into one model provider.
The Problem It’s Addressing
Cloud-hosted AI agents have a compliance problem. Most tools — Claude Code, Cursor, Codex CLI, GitHub Copilot — route your code and prompts through the vendor’s infrastructure to reach the model. For regulated industries, government contractors, and teams under strict data residency rules, that’s a blocker.
Coder’s own research found that 70% of companies are deploying agents on infrastructure that isn’t appropriate for their security requirements. CEO Rob Whiteley put it directly: “Companies are being forced to choose between adopting AI agents and maintaining control over their infrastructure and data. Coder Agents removes that tradeoff.”
Coder Agents runs in air-gapped and network-restricted environments. You choose the model, you control the traffic, the agent stays inside your perimeter.
How It Works
Coder Agents is native to the Coder control plane — it’s not a third-party agent you drop into a workspace. It provisions workspaces on demand when execution is needed, then spins them down. Developers interact with it through a conversational UI or API.
Tasks it can handle: writing code, generating tests, analyzing repositories, opening pull requests. For more complex workflows, it supports skills, MCP (Model Context Protocol), and subagents. It integrates with CI/CD pipelines, GitHub Actions, and Slack via API.
The agent separates execution from model selection. Switching from one provider to another doesn’t require rebuilding your workflow. That’s deliberate — Coder is betting that enterprise teams will want to swap models as pricing and capability evolve, and they don’t want to be re-architecting every time.
Beta Terms and What Comes Next
Full Premium features with no usage-based limits are available through September 2026 for beta participants. After that, pricing hasn’t been detailed publicly.
Coder Agents is intended to replace Coder Tasks, the existing task system. That migration won’t be forced immediately — the announcement committed to clear migration paths before any removal. The API is described as still maturing, with potential breaking changes possible before general availability.
Context
This enters a market that’s splitting in two directions: cloud-native agents (Claude Code, Codex CLI, GitHub Copilot cloud agent) and self-hosted alternatives for organizations that can’t use them. Kiro is Amazon’s version of the self-hosted play, though it’s tied to AWS Bedrock. Coder’s differentiator is the model-agnostic angle combined with the existing Coder Platform footprint — teams already running remote development environments on Coder can add agents without changing much.
Sources: Coder Blog — Introducing Coder Agents, GlobeNewswire press release (May 6, 2026), SD Times (May 8, 2026)