Example: Integrating OpenAI Codex with TensorZero

This example shows how to use OpenAI Codex with TensorZero — fully open-source and self-hosted.

Why?

Use every major model provider with Codex: Anthropic, AWS Bedrock, AWS SageMaker, Azure, DeepSeek, Fireworks, GCP Vertex AI Anthropic, GCP Vertex AI Gemini, Google AI Studio (Gemini API), Hyperbolic, Mistral, OpenAI, Together, vLLM, xAI, and any OpenAI-compatible provider (e.g. Ollama)
Gain comprehensive observability into your Codex usage, including detailed logs for every LLM call.
Set up advanced inference features like retries, fallbacks, load balancing, inference-time optimizations, experimentation (A/B testing), and more.
Collect data for fine-tuning and other optimization techniques, and use TensorZero recipes to create custom models for your Codex usage.

Getting Started

Prerequisites

Install Docker.
Install Node 24.13.0.
Generate credentials for the providers you want to use (e.g. ANTHROPIC_API_KEY).

Setup

Create a .env file with your provider credentials. (See .env.example for reference.)
Run docker compose up to start TensorZero.
Install Codex: npm i -g @openai/codex

Add the TensorZero Gateway to your Codex configuration (~/.config/config.yaml or ~/.codex/config.json):

yaml

model: "tensorzero::model_name::anthropic::claude-sonnet-4-5-20250929"
provider: tensorzero
providers:
  tensorzero:
    name: TensorZero
    baseURL: http://localhost:3000/openai/v1
    envKey: TENSORZERO_API_KEY # not used but required by Codex
  # ... other providers ...

json

{
  "model": "tensorzero::model_name::anthropic::claude-sonnet-4-5-20250929",
  "provider": "tensorzero",
  "providers": {
    "tensorzero": {
      "name": "TensorZero",
      "baseURL": "http://localhost:3000/openai/v1",
      "envKey": "TENSORZERO_API_KEY"
    }
  }
}

Run Codex with TensorZero:

bash

TENSORZERO_API_KEY="not-used" codex
# or set the environment variable in your shell and just run `codex`

You can replace tensorzero::model_name::anthropic::claude-sonnet-4-5-20250929 with any other model supported by TensorZero, e.g. tensorzero::model_name::mistral::open-mistral-nemo-2407.

You can also define custom TensorZero functions in the config/tensorzero.toml file, and use them with Codex as tensorzero::function_name::your_function_name. This will enable you to use advanced inference features, collect data for fine-tuning and other optimization recipes, and more. See our Quick Start Guide for more details.

<p align="center"> <em>OpenAI Codex Observability with the TensorZero UI</em> — <code>http://localhost:4000/</code></p>