Give a hand-set LLM provider its own default model

[thesanogoeffect]

Requested by Jakub · project thread

Before: LLM_PROVIDER=anthropic (or --llm-provider anthropic) without LLM_MODEL sent the Gemini default, gemini-3.5-flash-lite, to Anthropic, which answered 404. Every non-Google provider had the same problem. bibr setup always writes LLM_MODEL, so this hit hand-written envs, such as a Docker deploy.

After: an unset LLM_MODEL follows the provider, using the models bibr setup already writes: claude-haiku-5-5 for anthropic, gpt-5-nano for openai, llama-3.3-70b-versatile for groq, and gpt-oss:20b for ollama. An LLM_MODEL you set yourself is always kept. An openai provider with LLM_BASE_URL keeps the old default, since bibr can't guess what model a self-hosted server serves.

How:

• bibr/config.py: a new DEFAULT_LLM_MODELS table and an LlmOptions validator that fills model from it when model isn't in model_fields_set. The default is kept out of model_fields_set, so a later provider change still treats the model as unset. setup_wizard.LLM_DEFAULTS now reads its models from the same table.
• bibr/local/cli/run_config.py: --llm-provider swaps the provider after the settings are built, so it calls the same step again. Presets need nothing extra, because they rebuild the settings through the validators.
• Docs (configuration and deployment guides, .env.example) and the CHANGELOG describe the new default.
• Tests cover every provider, an explicit model being kept, the openai + LLM_BASE_URL case, a provider swap, and the CLI flag. The Anthropic adapter tests in test_provider_temperature.py and test_anthropic_thinking_caps.py now pin claude-haiku-4-5. Without a model, they would get Haiku 5.5, which rejects temperature and budget thinking, the very parameters those tests check.

Checks: ruff check, ruff format --check, and pytest -m "not slow" pass (14,213 passed). The only other failure was test_rate_limiter.py::test_serial_waiters_are_pacing, a timing test that fails under full-suite load and passes on its own, as noted on #207.

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.56%. Comparing base (d2b75ba) to head (15b99dc).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #210   +/-   ##
=======================================
  Coverage   90.55%   90.56%           
=======================================
  Files         309      309           
  Lines       57289    57305   +16     
  Branches    10376    10379    +3     
=======================================
+ Hits        51879    51897   +18     
+ Misses       3450     3449    -1     
+ Partials     1960     1959    -1     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.