fix: use resolved LLM key for PageIndex

[wachiravit-thitagran]

Summary

• pass the per-request credential bundle into long-document PageIndex indexing
• forward LLM_API_KEY, base URL, headers, and timeout through PageIndex llm_params
• use a narrow local-client adapter so an explicit resolved key satisfies provider validation without mutating process-wide OPENAI_API_KEY
• preserve the existing UI persistence and once-per-request credential resolution flow

Root cause

PageIndex validates provider-specific process environment variables during local client construction, before its per-call llm_params are used. OpenKB stores the UI key as LLM_API_KEY and resolves it into an isolated request bundle, so OpenAI-compatible models failed with OPENAI_API_KEY missing even though the UI key was configured.

Validation

• pytest tests/test_indexer.py tests/test_add_command.py tests/test_api.py -q (224 passed)
• ruff check and ruff format --check on changed files
• mypy openkb/cli.py openkb/indexer.py
• git diff --check

Warning

AI-generated code: This fix was generated by GPT-5.6 Sol.

[mike-011]

It looks like this is a related issue #205

[veereshkumar-se]

Formations · PR #209 · 820b989359cc · NO ACTION

— PR tests

— Intent

— Severity

0 Impacted

— Selfhealed

---

_Details

1 · PR tests

No PR test run recorded for this head.

2 · Classification

Intent: — · Severity: —

[wachiravit-thitagran]

Closing this PR as superseded by #250. The underlying PageIndex credential-forwarding issue is still relevant, but #250 covers it more comprehensively and also addresses the local SQLite connection cleanup problem (#249).