Use CURSOR_API_KEY for LLM narrative env config

[trenchsheikh]

Summary

Rename the optional LLM-narrative env vars to use a Cursor-style naming, and make the upstream endpoint configurable so a Cursor API key (or any OpenAI-compatible provider key) can be used.

Env var changes

Before	After
OPENAI_API_KEY	CURSOR_API_KEY
OPENAI_MODEL	CURSOR_MODEL
(new)	CURSOR_API_BASE_URL (default https://api.openai.com/v1)

What changed

• app/core/config.py: openai_api_key / openai_model settings replaced with cursor_api_key, cursor_model, plus a new cursor_api_base_url.
• app/services/underwriting/llm.py: builds the chat-completions URL from cursor_api_base_url and authenticates with cursor_api_key. Behavior is unchanged when the key is unset (rule-based narrative is returned and the LLM never affects outcome / amount / APR).
• .env.example and backend/.env.example: updated to the new variable names with explanatory comments.
• README.md: feature blurb, underwriting section, and config table updated to reference CURSOR_API_KEY / CURSOR_API_BASE_URL / CURSOR_MODEL.

Testing

• pytest -q — all 15 tests pass.
• Smoke test against a local mock HTTP server confirms:
  • request goes to ${CURSOR_API_BASE_URL}/chat/completions
  • Authorization: Bearer ${CURSOR_API_KEY} is sent
  • model field uses CURSOR_MODEL
  • narrative is enriched with the assistant note on success

Notes

CURSOR_API_BASE_URL defaults to OpenAI's host so existing deployments continue to work after just renaming the env var. Point it at any other OpenAI-compatible chat-completions endpoint to use a different provider.

Slack Thread