Model Router (deleted)
Status: REMOVED 2026-05-08.services/model_router.py + its companions (usage_tracker.py, model_constants.py) were deleted in the Phase 2 cleanup once LiteLLM had been on prod long enough to validate the cutover.
For the equivalent functionality today:
- Per-call provider routing + cost tracking + retries:
services/litellm_provider.md— theLiteLLMProviderplugin is the primary LLM router on prod for all four cost tiers (plugin.llm_provider.primary.{free,budget,standard,premium}='litellm'). - Model selection:
../cost-tier-routing.md— each step reads its own*_modelapp_settingspin (e.g.pipeline_writer_model,pipeline_critic_model); thecost_tier.<tier>.modelindirection was removed in PR #1907. - Per-1K-token cost lookups:
services/cost_lookup.py— wrapslitellm.model_cost.