Skip to main content

OpenAIIntegration

What it is​

  • A small integration wrapper around the openai Python client.
  • Adds:
    • Filesystem caching (1 day TTL) for API calls.
    • JSON persistence of retrieved models and chat completions to a configured datastore via StorageUtils and an ObjectStorageService.
    • Optional conversion into LangChain StructuredTool tools via as_tools().

Public API​

OpenAIIntegrationConfiguration​

Dataclass configuration for the integration.

  • Fields:
    • api_key: str — OpenAI API key.
    • datastore_path: str — Base path used to build output directories for saved JSON.
    • object_storage: ObjectStorageService — Storage backend used by StorageUtils.

OpenAIIntegration​

Integration class for OpenAI endpoints.

  • list_models() -> Dict

    • Calls client.models.list().
    • Saves models.json under: {datastore_path}/models/_all/models.json.
    • Returns: {"models": [ ...model dicts... ]}.
    • Cached for 1 day.

  • retrieve_model(model_id: str) -> Dict

    • Calls client.models.retrieve(model_id).
    • Saves {model_id}_info.json under: {datastore_path}/models/{model_id}/.
    • Returns: model dict.
    • Cached for 1 day.

  • create_chat_completion(prompt: Optional[str] = None, system_prompt: str = "You are a helpful assistant.", messages: List[Dict[str, str]] = [], model: str = "o3-mini", temperature: float = 0.3) -> Dict

    • Calls client.chat.completions.create(...).
    • If messages is empty and prompt is provided, constructs messages:
      • {"role": "developer", "content": system_prompt}
      • {"role": "user", "content": prompt}
    • If model starts with "o", it does not pass temperature; otherwise it does.
    • On success, saves the completion JSON under:
      • {datastore_path}/completions/{model}/{model}_{temperature}.json
    • Returns: {"content": <first choice message content>} or {} if content is unavailable.
    • Cached for 1 day.

as_tools(configuration: OpenAIIntegrationConfiguration) -> List[StructuredTool]​

Converts the integration into LangChain tools:

  • openai_list_models
  • openai_retrieve_model (args: model_id)
  • openai_create_chat_completion (args: prompt, system_prompt, model, temperature)

Configuration/Dependencies​

  • External dependencies:
    • openai (OpenAI client)
    • langchain_core.tools.StructuredTool
    • pydantic (tool argument schemas)
    • naas_abi_core:
      • Integration, IntegrationConfiguration
      • CacheFactory + DataType (filesystem cache)
      • ObjectStorageService, StorageUtils (JSON persistence)
  • Caching:
    • Uses CacheFactory.CacheFS_find_storage(subpath="openai").
    • All decorated methods have a TTL of 1 day and cache as DataType.PICKLE.

Usage​

from naas_abi_core.services.object_storage.ObjectStorageService import ObjectStorageService
from naas_abi_marketplace.ai.chatgpt.integrations.OpenAIIntegration import (
    OpenAIIntegration,
    OpenAIIntegrationConfiguration,
)

# You must provide an ObjectStorageService instance appropriate for your environment.
object_storage = ObjectStorageService(...)  # implementation-specific

cfg = OpenAIIntegrationConfiguration(
    api_key="YOUR_OPENAI_API_KEY",
    datastore_path="/tmp/naas/openai",
    object_storage=object_storage,
)

client = OpenAIIntegration(cfg)

# List models (cached; also writes models.json)
models = client.list_models()

# Retrieve one model (cached; also writes {model_id}_info.json)
model_info = client.retrieve_model("gpt-4.1-mini")

# Create a chat completion (cached; also writes completion JSON)
resp = client.create_chat_completion(
    prompt="Write a haiku about caching.",
    system_prompt="You are a concise assistant.",
    model="o3-mini",
    temperature=0.3,
)
print(resp.get("content"))

Using as LangChain tools:

from naas_abi_marketplace.ai.chatgpt.integrations.OpenAIIntegration import as_tools, OpenAIIntegrationConfiguration

tools = as_tools(cfg)
# tools is a list of StructuredTool objects usable in LangChain.

Caveats​

  • messages default is a mutable list ([]); if mutated externally it could affect subsequent calls.
  • The cache key for create_chat_completion includes messages and prompts stringified; large/complex inputs may produce unwieldy keys.
  • For models whose name starts with "o", temperature is not passed to the OpenAI API call (but is still used in cache key and filename).