v0.1.10

Prompt caching + auto model routing

• OpenAI: prompt_cache_key + 24h retention for improved cache hits
• Gemini: surface implicit caching (cachedContentTokenCount) in usage events
• Anthropic: documented 1h TTL option, normalized usage fields
• Cost tracking: cache-aware pricing across all providers (input/output/cache_read/cache_write)
• Auto model: smart routing — cheap model for tool turns, full model for responses (anthropic:auto, openai:auto, gemini:auto)
• SmartProvider: cross-provider fallback on failure
• Conversations DataView: per-model pricing instead of hardcoded Sonnet rates
• Fix: auto model resolution falls back to standard tier when no read tier exists