Bug: Invalid default `retrieve_model` value in config.yaml causes runtime error

[ChiragB254]

Description

The default configuration in pageindex/config.yaml sets retrieve_model to gpt-5.4, which is not a valid OpenAI model name. This causes a runtime error for any user relying on the default config when retrieve_model is used.

Affected File

pageindex/config.yaml — line 3

retrieve_model: "gpt-5.4"  # ← this model does not exist

Steps to Reproduce

1. Clone the repo and install dependencies
2. Set a valid OPENAI_API_KEY
3. Run PageIndexClient without overriding retrieve_model
4. Any call that routes through the retrieve model will fail with a model-not-found error from the LiteLLM/OpenAI API

Expected Behaviour

The default retrieve_model should be a valid, existing model name (e.g. gpt-4o or gpt-4.1), or should fall back to the value of model when not explicitly set.

Actual Behaviour

LiteLLM / OpenAI API returns an error because gpt-5.4 is not a recognised model identifier.

Proposed Fix

Update config.yaml to use a valid model name:

retrieve_model: "gpt-4o"

Alternatively, simply removing the retrieve_model key from config.yaml (making it None by default) would also fix this cleanly, since client.py already handles the fallback:

self.retrieve_model = _normalize_retrieve_model(opt.retrieve_model or self.model)

Environment

• Affects all users using the default config
• No special environment required to reproduce

---

I would like to work on this fix if the maintainers are open to it. Happy to submit a PR once this issue is confirmed. 🙂

[ChiragB254]

Closing this — gpt-5.4 is actually a valid OpenAI model. Filed in error.