Skip to main content

ArtificialAnalysisWorkflow

What it is​

A Workflow that:

  • Fetches model data from the Artificial Analysis API (LLMs or media endpoints).
  • Optionally filters the results to only include providers that have corresponding local agent modules.
  • Persists the (optionally filtered) response as a timestamped JSON file under storage/datastore/core/modules/abi/ArtificialAnalysisWorkflow.

Public API​

Classes​

  • ArtificialAnalysisWorkflowConfiguration(WorkflowConfiguration)

    • Holds API connection settings.
    • Fields:
      • api_key: str (required)
      • base_url: str (default: https://artificialanalysis.ai/api/v2)

  • ArtificialAnalysisWorkflowParameters(WorkflowParameters)

    • Input parameters for execution.
    • Fields:
      • endpoint: str (default: "llms") — endpoint to fetch (e.g., "llms", "text-to-image", etc.)
      • include_categories: bool (default: False) — adds include_categories=true query param for media endpoints
      • validate_agents_only: bool (default: True) — if True, filters models to only those with local agent modules

  • ArtificialAnalysisWorkflow(Workflow)

    • Main workflow implementation.

Methods​

  • ArtificialAnalysisWorkflow.run_workflow(parameters: ArtificialAnalysisWorkflowParameters) -> dict

    • Executes fetch → optional filter → save JSON.
    • Returns a dict including:
      • status ("success" or "error")
      • endpoint, models_count, original_count, valid_modules, output_file, timestamp (on success)

  • ArtificialAnalysisWorkflow.as_tools() -> list[BaseTool]

    • Exposes the workflow as a LangChain StructuredTool named artificial_analysis_data_fetch.
    • Tool calls into self.run(ArtificialAnalysisWorkflowParameters(**kwargs)).

  • ArtificialAnalysisWorkflow.as_api(...) -> None

    • Present but not implemented (returns None).

Configuration/Dependencies​

  • Network:
    • Uses requests.get with header x-api-key: <api_key>.
  • Filesystem:
    • Writes JSON files to: storage/datastore/core/modules/abi/ArtificialAnalysisWorkflow/.
    • When validate_agents_only=True, scans local modules under: src/core/modules/*/agents/*Agent.py.
  • Key dependencies:
    • requests
    • pydantic (for Field / schema)
    • langchain_core.tools (for StructuredTool)
    • naas_abi_core.workflow base classes

Usage​

Minimal direct invocation:

from naas_abi.workflows.ArtificialAnalysisWorkflow import (
    ArtificialAnalysisWorkflow,
    ArtificialAnalysisWorkflowConfiguration,
    ArtificialAnalysisWorkflowParameters,
)

wf = ArtificialAnalysisWorkflow(
    ArtificialAnalysisWorkflowConfiguration(api_key="YOUR_ARTIFICIAL_ANALYSIS_API_KEY")
)

result = wf.run_workflow(
    ArtificialAnalysisWorkflowParameters(
        endpoint="llms",
        include_categories=False,
        validate_agents_only=True,
    )
)

print(result["status"], result.get("output_file"))

Using as a LangChain tool:

wf = ArtificialAnalysisWorkflow(
    ArtificialAnalysisWorkflowConfiguration(api_key="YOUR_ARTIFICIAL_ANALYSIS_API_KEY")
)

tool = wf.as_tools()[0]
out = tool.invoke({"endpoint": "llms", "validate_agents_only": False})
print(out)

Caveats​

  • validate_agents_only=True depends on the local repository layout:
    • If src/core/modules does not exist, filtering will effectively keep all models (no valid modules found).
  • as_api(...) is a stub and does not register any routes.
  • The workflow prints progress and filtering details to stdout.