OpenAIAgentsCacheAdapter

Cache wrapper for the OpenAI Agents SDK. Intercepts Runner.run() and Runner.run_async() and returns cached results for identical agent + input combinations.

Installation

pip install openai-agents

---

Usage

from agents import Agent, Runner
from chengeta_ai import CacheManager, InMemoryBackend, CacheKeyBuilder
from chengeta_ai.adapters.openai_agents_adapter import OpenAIAgentsCacheAdapter

agent = Agent(
    name="assistant",
    instructions="You are a helpful assistant",
    model="gpt-4o",
)

manager = CacheManager(
    backend=InMemoryBackend(),
    key_builder=CacheKeyBuilder(namespace="myapp"),
)
adapter = OpenAIAgentsCacheAdapter(manager)

# Sync — wraps Runner.run_sync()
result = adapter.run(agent, "What is the capital of France?")

# Async — wraps Runner.run()
result = await adapter.arun(agent, "What is the capital of France?")
print(result.final_output)

With Disk Backend (persistent across restarts)

from chengeta_ai import DiskBackend

manager = CacheManager(
    backend=DiskBackend(directory="/var/cache/agents"),
    key_builder=CacheKeyBuilder(namespace="prod"),
)
adapter = OpenAIAgentsCacheAdapter(manager)

Invalidate by agent

adapter.invalidate_agent(agent)  # clears all cached results for this agent

---

How It Works

Cache key = hash(agent.name + agent.model + input_text + kwargs). The full RunResult object is serialised with pickle and replayed on cache hit.

Prompt caching is automatic in the OpenAI Agents SDK for repeated system prompts — OpenAIAgentsCacheAdapter adds an application-level cache on top, eliminating entire agent runs for known inputs.

---

API Reference

Method	Description
run(agent, input_text, **kwargs)	Cached sync Runner.run_sync()
arun(agent, input_text, **kwargs)	Cached async Runner.run()
invalidate_agent(agent)	Invalidate all cached results for an agent