Who it is forA contributor who wants to change the gateway, test recovery behavior, or submit a PR to the project.
The problemTesting failover fixes requires running the same launcher used by the package and Docker image, not only the container.
The outcomeA local source build of the gateway running with the same launcher as the release image.
What you need first
- Docker installed and running, or uv installed for running from source
- Upstream ChatGPT accounts you are authorized to use, to add in the dashboard
- A browser to open the dashboard at localhost:2455
- Codex CLI, desktop, or IDE integration, or another OpenAI-compatible client, if you plan to connect one
Steps
- Clone the repository from https://github.com/aafqaq/codex-lb-enhanced and change into its directory.
- Install dependencies:
uv sync --dev --frozen - Start the application launcher:
uv run codex-lb - Open http://localhost:2455 and complete initialization.
- Reproduce a failover case, then open a PR if you have a fix, as the README invites.
Prompt for your agent
In a clone of aafqaq/codex-lb-enhanced at <path>, run `uv sync --dev --frozen` and then `uv run codex-lb`. Report whether the launcher starts and whether http://localhost:2455 responds. Verify that no dependency install step failed before you start the launcher.
Caveat. The README notes that account-pool switching is still being improved and can occasionally fail to recover, so expect edge cases while you develop.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for aafqaq/codex-lb-enhanced and "Run the gateway from source for local development"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer