omnicache-ai v0.3.0

What's New in v0.3.0

New Adapters (4)

• GoogleADKCacheAdapter — Google ADK Agent.run() / run_async() — pip install 'omnicache-ai[google-adk]'
• OpenAIAgentsCacheAdapter — OpenAI Agents SDK Runner.run() — pip install openai-agents
• LlamaIndexLLMCacheAdapter + LlamaIndexQueryCacheAdapter — LlamaIndex LLM + QueryEngine — pip install 'omnicache-ai[llamaindex]'
• ClaudeAgentCacheAdapter — wraps claude_code_sdk.query() async generator — pip install claude-code-sdk

New Vector Backends (2)

• QdrantBackend — fastest vector DB in 2026 (22ms p95), in-memory + remote — [vector-qdrant]
• WeaviateBackend — native hybrid search (vector + BM25) — [vector-weaviate]

New Cache Layers

• PromptCacheLayer — auto-injects Anthropic cache_control, tracks provider_cache_hits + estimated_cost_saved_usd
• AdaptiveSemanticCache — self-tuning threshold + max_turn_count multi-turn guard

New Core Utilities

• RequestConfig — per-request TTL / threshold / skip_cache overrides
• CacheWarmer — bulk warm from query list or CSV
• CacheManager.for_tenant(id) — per-tenant key namespacing, shared backend
• PrometheusExporter + OpenTelemetryExporter — production observability — [observability]
• Extended CacheMetrics with provider_cache_hits, estimated_tokens_saved, estimated_cost_saved_usd

Docs & Cookbook

• 11 new website doc pages (adapter, backend, layer, observability docs)
• 5 new cookbook examples

Install

pip install omnicache-ai==0.3.0

See CHANGELOG for full details.