Please verify openai-codex behavior in self-hosted Docker and document limitations

[Raimiz]

Hi Hindsight team!

Could you please verify this behavior on your side:

• Environment: self-hosted Hindsight in Docker (local dev/test)
• Provider: HINDSIGHT_API_LLM_PROVIDER=openai-codex
• Auth source: ~/.codex/auth.json (Codex CLI OAuth)
• Models tested: gpt-5.3-codex, gpt-5.1-codex-mini

What we see:

• Codex CLI works interactively on host
• Hindsight programmatic calls fail with 429 usage_limit_reached
• Error details include plan_type=free
• Endpoint in logs: https://chatgpt.com/backend-api/codex/responses

If this is expected/unsupported for self-hosted programmatic usage, please mark it explicitly in docs (and ideally add a startup warning), so users don’t spend time debugging this path.

If useful, I can share sanitized logs and compose config.