Modly extension for Codex image workflows, AI image generation, and image editing inside Modly.
Experimental Modly model extension that lets Modly generate local images through a locally installed, locally authenticated Codex runtime.
In user terms: after the extension is installed in Modly and Codex is already working on the host, Modly can expose a Codex Local Image Model for: Text-to-image: enter a prompt and receive one saved local image. Image-to-image: provide one primary image plus a prompt and receive one saved local image, with up to three optional reference images routed by Modly's multi-image contract.
The extension returns a single absolute local image path. On the currently validated local Modly host path, that image is intended to be preview-compatible through Modly's image output handling.
This project is an independent integration experiment and is not affiliated with, endorsed by, or sponsored by OpenAI. Codex, ChatGPT, and OpenAI are third-party products, services, and marks belonging to their respective owners. Use of any local Codex runtime, ChatGPT entitlement, or related OpenAI service remains subject to the applicable OpenAI terms and product availability.
Current V1 support state Extension version: 0.1.4 Support state: experimental Modly surface owner: FastAPI model extension Bucket: model-managed-setup Implementation profile: python-local-bridge Setup contract: user-managed Codex login; extension-managed Python venv, official openai-codex==0.154.0 SDK with its matching openai-codex-cli-bin==0.154.0, and host preflight checks Headless eligibility: conditional; generation can run through Modly's backend model surfaces, but GitHub install/repair and app-level flows remain outside this extension's headless contract Historical host evidence: the previous legacy integration generated on linux/arm64 with codex-cli 0.122.0, passed readiness with 0.124.0, and was smoke-tested once on windows/x8664; this evidence does not validate the new official SDK path Current migration status: E2E PASS on the validated Linux ARM64 host with openai-codex==0.154.0 and openai-codex-cli-bin==0.154.0. Runtime setup, health/reload, default-model generation, multi-input generation with explicit gpt-6-astra, workspace persistence, HTTP image serving, and visual QA all passed Codex CLI compatibility policy: minimum >= 0.122.0 by default; newer versions pass preflight as experimental/unvalidated until smoke evidence records them Configured platform allowlist: darwin/arm64, darwin/x8664, linux/arm64, linux/x8664, and windows/x8664; allowlisting is not E2E validation Linux ARM64 risk: still marked high in metadata because it is validated on the current host path, not proven as a broad portability guarantee Official package evidence: PyPI publishes openai-codex==0.154.0, pins openai-codex-cli-bin==0.154.0, and provides a manylinux217aarch64 runtime wheel; the SDK wheel and public signatures were inspected in a temporary ARM64 environment
Read the full README on GitHub →