Anthropic has released self-hosted Claude Code environments in public beta, allowing enterprise and team-plan customers to run AI coding agent sessions directly on their own infrastructure rather than relying on Anthropic's cloud servers. This move addresses a critical barrier for compliance-sensitive organizations in finance, healthcare, and defense that previously couldn't adopt Claude Code due to data residency requirements or network isolation policies.

What Problem Does Self-Hosted Claude Code Solve?

For years, organizations handling sensitive data have faced a dilemma with AI coding tools: the models work well, but the data has to leave your network to be processed on someone else's servers. Even if you opt out of training data usage, the information still travels outside your firewall, which can violate compliance requirements or internal security policies. Self-hosted Claude Code eliminates that friction by keeping agent execution inside your own network, next to your existing services, toolchains, and security controls.

This is particularly significant for industries with strict regulatory requirements. Financial services firms, healthcare organizations, and defense contractors have been waiting for exactly this capability. The feature is now available to Team and Enterprise plan customers, though it's off by default and requires explicit setup.

How Does Self-Hosted Claude Code Actually Work?

Anthropic offers two operational modes for self-hosted environments, each designed for different scaling needs:

• Fixed Mode: Maintains a set number of runners with sessions distributed across them, providing predictable resource allocation and consistent performance for steady workloads.
• On-Demand Mode: Watches for queued sessions, spins up a runner when work arrives, and shuts it down when finished, so capacity scales directly with demand without paying for idle infrastructure.

This flexibility means organizations can choose the deployment model that matches their infrastructure and budget constraints. On-demand mode is particularly attractive for teams with variable coding workloads, while fixed mode suits organizations with consistent usage patterns.

What Else Changed in Claude Code's Enterprise Tooling?

Beyond self-hosted environments, Anthropic expanded its compliance infrastructure significantly. The Claude Compliance API now covers Cowork, Claude Code, and the command-line interface (CLI) across desktop, web, mobile, and other surfaces, giving security teams a single API to pull unified session content and metadata for audits and electronic discovery. This unified audit trail is critical for organizations that need to demonstrate to legal or information security teams that all AI-assisted work is logged and reviewable.

When paired with self-hosted environments, this creates a configuration that can satisfy even the strictest enterprise security reviews: air-gapped execution plus full audit trails. Organizations can now run Claude Code entirely within their network while maintaining complete visibility into every session.

Additionally, Claude Code received a major bug-fix release on August 12 that addressed session redraw failures, Git detection issues on Windows, memory leaks in Remote Control resume functionality, and self-hosted runner reliability improvements. For teams already running Claude Code daily, particularly on Windows or with self-hosted runners, this release directly addresses known pain points.

How Does This Fit Into the Broader AI Coding Landscape?

Claude Code's self-hosted option arrives as AI coding harnesses become increasingly important infrastructure. A harness is middleware that handles input passed to an AI model and the output returned from it, overseeing prompts, context management, tool orchestration, and safety concerns. Claude Code serves as Anthropic's harness for its Claude model family, similar to how Codex functions for OpenAI's GPT models.

The competition in AI coding is shifting from model performance alone to harness design and deployment flexibility. DeepSeek recently released an open-source harness that treats everything as a plugin, allowing developers to swap models, tools, and infrastructure components without changing source code. This modular approach is reshaping how developers think about AI coding infrastructure.

"I don't think the DeepSeek Harness is perfect but this is for sure the first time I have been looking at something new in the space and felt quite inspired to revisit some of our choices," said Armin Ronacher, co-founder of AI company Earendil.

Armin Ronacher, Co-founder of Earendil

Anthropic's self-hosted option represents a different competitive strategy: keeping the harness proprietary but giving enterprises control over where their code execution happens. This appeals to organizations that want Claude's capabilities without the data residency concerns.

Steps to Evaluate Self-Hosted Claude Code for Your Organization

• Assess Compliance Requirements: Review your organization's data residency, network isolation, and audit trail requirements to determine if self-hosted deployment is necessary for your use case.
• Evaluate Infrastructure Capacity: Determine whether fixed or on-demand runner modes align better with your existing infrastructure, budget, and coding workload patterns.
• Plan Audit Integration: Work with your security and legal teams to integrate the Claude Compliance API into your existing audit and eDiscovery workflows before deployment.
• Test with Pilot Team: Start with a small team or department to validate the self-hosted setup, runner performance, and audit logging before rolling out organization-wide.

The self-hosted beta is available now for Team and Enterprise plan customers, though it's disabled by default. Organizations interested in evaluating the feature should contact Anthropic to enable it and access the necessary documentation.