Configure the ChatGPT desktop app with Amazon Bedrock in AWS GovCloud (US)

Amazon Web Services and OpenAI have made the ChatGPT desktop application available on Amazon Bedrock in AWS GovCloud (US), routing model inference directly through us-gov-west-1 endpoints rather than standard commercial services. As detailed in AWS Machine Learning Blog's report, the architecture enables public sector teams to run OpenAI frontier models locally while binding all inference traffic to AWS Identity and Access Management credentials, managed workstation controls, and isolated cloud boundaries.

This deployment walkthrough covers the security parameters, system configurations, and identity requirements needed to stand up a locked-down instance across Windows, Linux, and macOS environments. It is designed specifically for government agencies, regulated enterprises, and security administrators who require strict model inference residency, zero data retention guarantees, and explicit network isolation before authorizing developer workstations for broader production use.

What shipped in Configure the ChatGPT desktop app with Amazon

The release introduces a controlled local deployment pattern that routes the ChatGPT desktop app and Codex developer interfaces directly to the Amazon Bedrock Mantle endpoint in us-gov-west-1. Users interact through two primary operating components: Work, which handles multi-step operational planning, file ingestion, research tasks, and review-ready analytical deliverables, and Codex, an agent tailored for software development tasks such as code editing, automated test execution, and pull request reviews. The setup supports the native ChatGPT desktop client, the Codex command-line interface, and integrated development environment extensions across Visual Studio Code, JetBrains, and Xcode through the Responses API on Amazon Bedrock.

The deployment structure strips standard commercial ChatGPT workspace controls and account roles out of the administrative path, replacing them with host operating system policies and AWS Identity and Access Management permissions. Authentication runs entirely through local device configuration files and the AWS SDK credential chain. Organizations can govern access through dedicated IAM roles, enterprise single sign-on mechanisms via AWS IAM Identity Center, or direct Bedrock bearer tokens stored in local environment paths, ensuring that every inference call adheres to defined AWS GovCloud boundaries.

What improved in Configure the ChatGPT desktop app with Amazon

The integration replaces commercial multi-tenant SaaS routing with localized AWS GovCloud network enforcement, per-token billing, and zero data retention compliance. In standard commercial setups, administrative policies rely on web workspace settings and platform SaaS rules, whereas this architecture enforces lockouts through administrator-owned device files. Furthermore, prompts and model completions processed under zero data retention mode are handled without being written to durable storage by AWS or transmitted back to OpenAI for model training or retention.

The change eliminates developer license commitments and recurring seat fees in favor of exact token usage billed against existing AWS accounts. Data residency is strictly confined to the chosen AWS GovCloud Region. Network traffic rules configured in the desktop client explicitly block connections to standard commercial domains, telemetry collection servers, and public update channels while leaving the designated Bedrock Mantle gateway as the sole allowable outbound path.

What you gain from Configure the ChatGPT desktop app with Amazon

Engineering organizations operating under strict regulatory compliance frameworks gain access to GPT-driven models without compromising institutional governance models. Because inference is processed through Amazon Bedrock in AWS GovCloud (US-West), teams subject to federal guidelines can utilize advanced coding automation while maintaining compliance with internal system security plans. Developers receive native agent capabilities to parse codebases and review pull requests without exposing proprietary source code to shared external services.

Administrators gain full architectural auditing and perimeter defense over local AI clients. The administrator requirements file allows security operators to selectively disable high-risk client behaviors, including unrestricted web searching, local socket bindings, automated background application updates, browser-based remote control, and arbitrary external plugins. This granular control gives defense and public sector compliance teams an auditable framework where every network transaction maps back to an IAM identity and an isolated cloud endpoint.

How to get Configure the ChatGPT desktop app with Amazon

Deployment requires establishing pre-launch administrative security files and user configurations across host workstations. On Windows, the system administrator creates the rules file at %ProgramData%\OpenAI\Codex\requirements.toml, while Linux and macOS hosts place it at /etc/codex/requirements.toml. The user configuration is stored at %USERPROFILE%\.codex\config.toml on Windows or ~/.codex/config.toml on Linux and macOS, accompanied by the named AWS profile such as codex-il5 pointing to us-gov-west-1.

# Set bearer token environment path for Bedrock authentication (macOS/Linux)
echo 'AWS_BEARER_TOKEN_BEDROCK="your-token-here"' >> ~/.codex/.env

# User runtime configuration in ~/.codex/config.toml
model = "REPLACE_WITH_APPROVED_MODEL_ID"
model_provider = "amazon-bedrock"

# Verify AWS GovCloud named profile credentials via AWS CLI
aws sts get-caller-identity --profile codex-il5

# Admin requirements baseline in /etc/codex/requirements.toml
allowed_login_methods = ["api"]
allowed_sandbox_modes = ["read-only", "workspace-write"]
allow_browser_and_computer_use = false

System administrators must complete credential verification and write the configuration files before launching the desktop client or the Codex command-line interface.

What to watch after Configure the ChatGPT desktop app with Amazon

Deploying the ChatGPT desktop app with AWS GovCloud does not provide blanket accreditation for production workloads. Each deploying organization remains fully responsible for conducting its own technical review and authorization process, validating the selected model ID, IAM identity policies, workstation controls, source repositories, and local data flows against its established system security plan. Security teams must ensure that credential helper tools, single sign-on token caches, and local AWS config files are write-protected from local desktop tasks.

Administrators should also continuously inspect endpoint routing to ensure client updates or environment modifications do not bypass domain filters. The client must be monitored to confirm it routes strictly to bedrock-mantle.us-gov-west-1.api.aws while maintaining denial rules against external services like api.openai.com, chatgpt.com, telemetry collectors, and remote update endpoints. Maintaining strict permission boundaries on both the local host files and the IAM roles ensures the installation remains compliant with internal federal security mandates.