Set up OpenAI ChatGPT Codex with LiteLLM on Amazon ECS and Amazon Bedrock

OpenAI's ChatGPT Codex, integrated with LiteLLM, offers a centralized control point for enterprise generative AI coding agents. These agents assist developers in understanding repositories, writing code, running tests, and completing complex engineering tasks. As organizations transition from experimentation to managed deployment, consistent control over model access, attribution, budgets, and telemetry becomes essential.

The deployment architecture involves placing LiteLLM between Codex and Amazon Bedrock. LiteLLM acts as a shared control point for authentication, routing, budgets, rate limits, and telemetry, while Codex manages local task execution and tool running on the developer workstation. The request flow includes Codex sending task context to LiteLLM, which authenticates and routes the request to Amazon Bedrock, then returns the generated text or function call. If tools are needed, Codex executes them locally, and results are sent back through LiteLLM.

This setup leverages AWS services such as Amazon ECS on Fargate, Application Load Balancer, AWS WAF, Amazon RDS, Secrets Manager, CloudWatch, and ECR. The deployment process involves cloning the LiteLLM repository, configuring environment variables, building Docker images, deploying CloudFormation stacks, and provisioning scoped API keys. The system supports secure, scalable, and controlled access to models, with the ability to enforce budgets, rate limits, and scoped identities.

For enterprise use, organizations can define model policies, issue scoped keys, and control model access via LiteLLM. Alternatively, direct access via AWS IAM Identity Center is suitable when native AWS controls suffice, offering simpler integration without a centralized gateway. Managed solutions like Portkey provide hybrid or fully managed control planes, suitable for organizations preferring vendor-managed infrastructure.

The article emphasizes testing and validation of the deployment, including request logging, API contract compliance, and operational health checks. It also discusses cleanup procedures for decommissioning the infrastructure. Overall, routing Codex through LiteLLM enhances control, observability, and policy enforcement in enterprise AI deployments.