ypollak2/llm-router

What's New

Phase 5: Comprehensive Test Suite & Release

• Added 27 comprehensive tests for all v5.0-v5.3 features
• Covers Ollama discovery injection, live API enumeration (OpenAI/Gemini), always-on dynamic routing, and /score endpoint
• Removed obsolete feature flags from test suite
• Fixed all code quality issues (ruff linting)

Architecture Highlights

• Live discovery cache with 30-minute TTL
• Fallback-safe routing when discovery empty
• Graceful degradation on API failures
• Free-first model ordering (Ollama/Codex before paid APIs)
• Production-ready error handling and timeouts

Files Changed

• 16 files modified
• 1,041 insertions(+), 141 deletions(-)
• 27 new comprehensive tests in test_adaptive_router.py

Installation

Upgrade to the latest version:
```bash
pip install --upgrade claude-code-llm-router==5.4.0
llm-router install
```

Or via Claude Code:
```bash
claude plugin reinstall llm-router
```

What Changed Since v5.3.2

• Phase 1: Ollama discovery injection — cache hit with env var fallback
• Phase 2: Live API enumeration — OpenAI and Gemini models auto-discovered
• Phase 3: Always-on dynamic routing — feature flag removed, routing always adaptive
• Phase 4: Sidecar /score endpoint — HTTP API for model scoring/ranking
• Phase 5: Comprehensive test suite + release validation

Test Coverage

• 27 new adaptive router tests
• Backward compatibility verified
• All existing tests updated and passing
• Code quality: ruff linting 100% compliant