Coder announced on September 15 that Claude Code is now available on Agent Relay, its self-hosted execution environment for cloud coding agents. The practical meaning: financial services firms, defense contractors, and other regulated enterprises that have been locked out of the industry’s leading coding agent due to compliance requirements now have a direct path in. Anthropic runs the reasoning; every tool call executes on machines the customer controls.
The Wall That Kept Regulated Teams Out #
Claude Code is not a coding assistant that suggests a line here and there. It reads your codebase, plans changes across multiple files, runs tests, and iterates — autonomously. That scope of access is exactly what compliance teams in regulated industries could not sign off on when the execution happened on Anthropic’s infrastructure.
Financial services firms operate under a layered compliance stack: GLBA, PCI DSS, NYDFS Part 500, DORA, and GDPR, often simultaneously. None of those frameworks permit autonomous agents to access source code, credentials, and internal services on vendor-controlled infrastructure. The result was that the developers who arguably had the most to gain from Claude Code — those in the most complex, legacy-heavy, security-sensitive environments — were also the ones completely blocked from using it.
Coder president Josh Epstein did not bury the demand signal: “Claude Code is one of the agentic tools our enterprise customers ask for by name, and regulated industries have wanted it the most.”
How Agent Relay Splits the Problem #
Agent Relay solves this by separating the agent into two distinct layers. The reasoning layer stays in Anthropic’s cloud — Claude runs the agent loop, manages billing, serves the model. The execution layer runs inside Coder workspaces on the customer’s own infrastructure — every file read, every shell command, every credential access, every internal service call happens on machines the customer controls, never Anthropic’s.
Each session gets its own sandboxed, ephemeral workspace scoped to that Claude Code session. The workspace opens an outbound connection to Anthropic’s backend. Nothing flows inbound. Code, secrets, and internal services never leave the VPC. Cat Wu, Claude Code product head at Anthropic, explained the target customer clearly: “The teams that care most about running Claude Code on their own infrastructure tend to be the ones with the most rigorous environmental isolation requirements.”
What Platform Teams Actually Control #
This is not simply self-hosted Claude Code. It is Claude Code inside Coder’s full enterprise governance stack, and for platform teams that distinction matters considerably:
- Agent Firewall : Egress policies are set once at the environment level and enforced across every workspace. A bank’s platform team can restrict access to only their internal git host and artifact registry; everything else is blocked and logged by default.
- Audit logging : Every agent run produces a complete record of what was accessed, executed, changed, and blocked — without manual reconstruction, ready for any compliance reporting window.
- Identity integration : Agent Relay maps agent actions to the human who initiated them through the organization’s identity provider. RBAC policies apply equally to human developers and Claude Code sessions.
- Environment templates : Approved models, permitted data sources, and available resources are defined once at the environment level. Every workspace derived from that template inherits the same policy. No per-team reconfiguration required.
For teams already running Agent Relay with Cursor — which launched September 2 — turning on Claude Code requires no new deployment and no additional security review. The governance infrastructure is already in place; Claude Code inherits it on day one.
Early Access and How to Get In #
Claude Code on Agent Relay is in early access with select design partners as of September 15. Existing Coder enterprise customers can contact their account team or visit coder.com/anthropic to request access. Full documentation is available at coder.com/docs/ai-coder/agent-relay.
The Pattern This Establishes #
Coder launched Agent Relay with Cursor on September 2. Claude Code followed on September 15 — thirteen days later. The pace is deliberate. According to Coder, Gartner projects 40% of enterprise applications will feature task-specific AI agents by the end of 2026, while only 31% of organizations have achieved organization-wide AI governance. That gap — high demand, low governance coverage — is exactly the market Agent Relay is targeting.
The pattern is now established: every serious coding agent needs a regulated-industry relay story. Cursor and Claude Code have theirs. The enterprises that have been citing compliance as the reason they cannot adopt agentic development tools are running out of that excuse.