## biblio LLM backend: document Claude Max plan as primary use case The `biblio

biblio LLM backend: document Claude Max plan as primary use case

The biblio_extract and other LLM-dependent tools use claude -p (CLI print mode) as the default backend, which is correct for Claude Max plan users who don't want to burn API tokens. This is documented in llm.py but worth noting:

1. The CLI backend is the right default for Max plan users
2. --max-tokens was removed from the CLI call (not a valid CLI flag) — the CLI handles limits automatically
3. The API fallback (BIBLIO_LLM_BACKEND=api) exists for CI/headless servers with API keys
4. Consider documenting this in biblio README or biblio.yml config comments

Also: the _API_MODEL_MAP in llm.py should be kept up to date as new Claude models ship. Current mapping uses claude-sonnet-4-20250514 etc.

Related Notes

• idea-arash-20260409-210844-158219.md — Both concern biblio toolchain improvements; systematic lit processing depends on LLM backend reliability
• idea-arash-20260409-231800-246938.md — biblio_status dashboard would surface LLM backend health/configuration as part of pipeline completeness
• idea-arash-20260409-214919-989861.md — biblio MCP tool — same ecosystem; crossref resolution may also use LLM backend for enrichment
• idea-arash-20260326-155537-817950.md — biblio_docling async feature request is in the same LLM-dependent biblio toolchain being discussed
• idea-arash-20260409-231734-553212.md — biblio_ingest bridge tool is another LLM-dependent biblio tool affected by the CLI vs API backend choice