Docker image for running codex

[ChrisL-GU]

I read in the documentation it is recommended to run codex within Docker to sandbox it. The script to run is provided, codex-cli/scripts/run_in_container.sh, but when running it there's an error because it cannot find the Docker image. There isn't any information provided about the Docker image to use. Is the intention for users to clone the repo and use codex-cli/scripts/build_container.sh to generate the Docker image to use?

[shanness]

Same.. Frustrating

[COWI-MI]

codex is light, so might be better to choose main environment and add codex.
For instance use a Python 3.12 or a CUDA image, add npm and install codex, instead of using a codex image and add a CUDA invironment to that.

[p3nGu1nZz]

Codex Container Build Instructions

To set up your own Codex container, follow these steps:

1. Build Your Own

   • The Dockerfile is included in codex-cli.
   • You must manually fetch all Codex dependencies and build them locally.
   • Make sure to use a Node v22 or greater slim image in your Dockerfile

2. Dependency Handling

   • Building codex locally with pnpm install && pnpm build generates a tarball in /dist.
   • The build container script requires this tarball; without it, the process will fail.
   • CUDA is NOT required for the image.

3. Provider Architecture

   • Codex operates on a provider-based architecture and communicates with:
     • A local Responses-compatible LLM server (e.g., Ollama or Llama.cpp in server mode).
     • Cloud-based providers via API (e.g., OpenAI).
   • API keys must be set in .env before container build (or manually copied afterward).
   • Ollama does not require API keys for operation.

4. Environment & Compatibility

   • Successfully tested on Windows 11 Home with WSL Ubuntu Server.
   • Dockerfile Update Required:
     • Update codex-cli Dockerfile to Node 22-slim instead of Node 20-slim (Codex requires Node v22).