Disable default cache when both cache and cache_seed are not set

[kumaranvpl]

Why are these changes needed?

Related issue number

Closes #1640

Checks

• I've included any doc changes needed for https://docs.ag2.ai/. See https://docs.ag2.ai/docs/contributor-guide/documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[Copilot]

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

[marklysze]

Hey @kumaranvpl, thanks for putting this together.

This is working for me now:

# NOT CACHED - CORRECT
llm_config = LLMConfig(config_list=[{"i_type": "openai", "model": "gpt-4.1-mini-2025-04-14"}])

# NOT CACHED - CORRECT
llm_config = LLMConfig(config_list=[{"api_type": "openai", "model": "gpt-4.1-mini-2025-04-14"}], cache_seed=None)

# CACHED - CORRECT
llm_config = LLMConfig(config_list=[{"api_type": "openai", "model": "gpt-4.1-mini-2025-04-14"}], cache_seed=49)
# CACHED CHANGE SEED - CORRECT
llm_config = LLMConfig(config_list=[{"api_type": "openai", "model": "gpt-4.1-mini-2025-04-14"}], cache_seed=48)

When using the Cache context manager, I tried the below and it didn't work...

from autogen.cache import Cache

with Cache.disk(cache_seed=39, cache_path_root=".cache"):

    llm_config = LLMConfig(config_list=[{"api_type": "openai", "model": "gpt-4.1-mini-2025-04-14"}])

    my_agent = ConversableAgent(
        name="test_agent",
        llm_config=llm_config,
    )

    my_other_agent = ConversableAgent(
        name="other_agent",
        llm_config=llm_config,
    )

    result = my_agent.initiate_chat(
        recipient=my_other_agent,
        message="Why is the sun round?",
        max_turns=2
    )

    print(result.cost)

Second run output:

{
'usage_including_cached_inference': {'total_cost': 0, 'gpt-4.1-mini-2025-04-14': {'cost': 0, 'prompt_tokens': 343, 'completion_tokens': 244, 'total_tokens': 587}},
'usage_excluding_cached_inference': {'total_cost': 0, 'gpt-4.1-mini-2025-04-14': {'cost': 0, 'prompt_tokens': 343, 'completion_tokens': 244, 'total_tokens': 587}}
}

... then I realised I needed to pass that cache into the initiate_chat as well, like the below. Just wondering why this is necessary?

from autogen.cache import Cache

with Cache.disk(cache_seed=39, cache_path_root=".cache") as dc:

    llm_config = LLMConfig(config_list=[{"api_type": "openai", "model": "gpt-4.1-mini-2025-04-14"}])

    my_agent = ConversableAgent(
        name="test_agent",
        llm_config=llm_config,
    )

    my_other_agent = ConversableAgent(
        name="other_agent",
        llm_config=llm_config,
    )

    result = my_agent.initiate_chat(
        recipient=my_other_agent,
        message="Why is the sun round?",
        max_turns=2,
        cache=dc
    )

    print(result.cost)

[marklysze]

It's working as planned now, thank you! I'll approve the changes.

I think we need some specific documentation on caching, I'll add an Issue.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

❗ There is a different number of reports uploaded between BASE (f7a935e) and HEAD (4c077ea). Click for more details.

HEAD has 1453 uploads less than BASE

Flag	BASE (f7a935e)	HEAD (4c077ea)
3.13	95	0
macos-latest	115	0
commsagent-discord	9	0
optional-deps	175	0
core-without-llm	14	1
3.9	95	0
ubuntu-latest	170	1
3.10	108	0
3.12	36	0
3.11	76	1
browser-use	7	0
commsagent-slack	9	0
windows-latest	125	0
commsagent-telegram	9	0
jupyter-executor	9	0
retrievechat-pgvector	10	0
rag	7	0
retrievechat-qdrant	14	0
retrievechat	15	0
retrievechat-mongodb	10	0
graph-rag-falkor-db	6	0
docs	6	0
interop	13	0
crawl4ai	13	0
wikipedia-api	13	0
google-api	13	0
twilio	9	0
interop-pydantic-ai	9	0
websockets	9	0
interop-langchain	9	0
mcp	13	0
interop-crewai	9	0
mistral	14	0
agent-eval	1	0
gpt-assistant-agent	3	0
lmm	4	0
long-context	3	0
cohere	15	0
teachable	4	0
gemini	15	0
retrievechat-couchbase	3	0
websurfer	15	0
ollama	15	0
swarm	14	0
cerebras	15	0
groq	14	0
llama-index-agent	3	0
bedrock	15	0
anthropic	16	0
together	14	0
integration	24	0
falkordb	2	0
core-llm	9	0
neo4j	2	0
gemini-realtime	1	0
captainagent	1	0
autobuild	1	0
openai-realtime	1	0
deepseek	1	0
openai	1	0

Files with missing lines	Coverage Δ
autogen/oai/client.py	53.53% <100.00%> (-23.88%)	⬇️

... and 79 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[kumaranvpl]

@marklysze I agree with you regarding context manager. If we are using context manager, then initiate_chat should automatically use cache provided by context manager.