Adapters

OmniCache-AI adapters provide drop-in integrations with popular AI and agent frameworks. Each adapter implements the framework's native cache or agent interface, so you get caching without changing your existing code.

Overview

Adapters bridge the gap between OmniCache-AI's cache engine and the framework-specific APIs that each AI library expects. Instead of writing custom glue code, you instantiate an adapter with a CacheManager and plug it into the framework's standard extension point.

There are two adapter styles:

• Interface adapters (LangChain, LangGraph) -- subclass the framework's cache/checkpointer base class and implement its required methods. The framework calls these methods automatically.
• Wrapper adapters (AutoGen, CrewAI, Agno, A2A) -- wrap an agent or handler with cache logic. All non-overridden attributes proxy through to the original object via __getattr__.

---

Framework Support Matrix

Adapter	Framework	Min Version	Extra	Interface
LangChainCacheAdapter	LangChain	langchain-core >= 0.2	pip install 'omnicache-ai[langchain]'	BaseCache
LangGraphCacheAdapter	LangGraph	langgraph >= 0.1	pip install 'omnicache-ai[langgraph]'	BaseCheckpointSaver
AutoGenCacheAdapter	AutoGen	pyautogen >= 0.2 or autogen-agentchat >= 0.4	pip install 'omnicache-ai[autogen]'	Agent wrapper
CrewAICacheAdapter	CrewAI	crewai >= 0.28	pip install 'omnicache-ai[crewai]'	Crew wrapper
AgnoCacheAdapter	Agno	agno >= 0.1	pip install 'omnicache-ai[agno]'	Agent wrapper
A2ACacheAdapter	A2A (Agent-to-Agent)	--	pip install omnicache-ai	Handler wrapper / decorator

---

Quick Start

All adapters follow the same three-step pattern:

from omnicache_ai import CacheManager, InMemoryBackend, CacheKeyBuilder

# 1. Create a CacheManager
manager = CacheManager(
    backend=InMemoryBackend(),
    key_builder=CacheKeyBuilder(namespace="myapp"),
)

# 2. Import the adapter
from omnicache_ai.adapters.langchain_adapter import LangChainCacheAdapter

# 3. Plug it in
adapter = LangChainCacheAdapter(manager)

tip

Install the all extra to get every framework dependency at once:

pip install 'omnicache-ai[all]'

---

Adapter Architecture

All wrapper-style adapters (AutoGen, CrewAI, Agno, A2A) implement the transparent proxy pattern:

1. Cache-aware methods (run, kickoff, process, etc.) check the cache before delegating to the wrapped object.
2. All other attribute accesses are forwarded to the wrapped object via __getattr__, so the adapter behaves identically to the original object for any non-cached operations.

# The adapter acts like the original object
cached_agent = AutoGenCacheAdapter(agent, manager)
cached_agent.name          # proxied to agent.name
cached_agent.run("hello")  # cache-aware

---

Choosing the Right Adapter

If you use...	Use this adapter	Why
langchain LLMs / chat models	LangChainCacheAdapter	Implements BaseCache -- set it globally via set_llm_cache()
langgraph state graphs	LangGraphCacheAdapter	Implements BaseCheckpointSaver -- pass to compile(checkpointer=...)
pyautogen 0.2.x agents	AutoGenCacheAdapter	Wraps generate_reply() with caching
autogen-agentchat 0.4+ agents	AutoGenCacheAdapter	Wraps run() / arun() with caching
crewai crews	CrewAICacheAdapter	Wraps kickoff() / kickoff_async()
agno agents	AgnoCacheAdapter	Wraps run() / arun()
Custom A2A / inter-agent messaging	A2ACacheAdapter	Wraps any handler via process() or @wrap decorator
Custom LLM functions (no framework)	Middleware	Use LLMMiddleware or AsyncLLMMiddleware directly

---

Next Steps

• LangChain Adapter -- Global LLM cache for LangChain
• LangGraph Adapter -- Checkpoint persistence for state graphs
• AutoGen Adapter -- Cached agent replies for both API generations
• CrewAI Adapter -- Cached crew kickoff results
• Agno Adapter -- Cached agent runs
• A2A Adapter -- Cached inter-agent messaging