Pipeline Module

async agent.pipeline.pipeline.run_pipeline(task)[source]

Execute the end-to-end research pipeline and return structured output.

Stages:

1) Strategy: generate multiple queries for the task
2) Retrieval: collect candidates from arXiv
3) Ranking: score with BM25 over title+abstract
4) Analysis: LLM/heuristic analysis of top candidates
5) Decision: choose items and produce a report
Parameters:

task (PipelineTask) – Validated agent.pipeline.models.PipelineTask describing user intent.

Return type:

PipelineOutput

Returns:

Structured PipelineOutput with analyzed items and an optional human report.

agent.pipeline.pipeline.run_pipeline_sync(task)[source]

Run run_pipeline() synchronously.

This helper creates and runs an event loop to execute the async pipeline in simple scripts or REPLs.

Parameters:

task (PipelineTask) – The pipeline task to execute.

Return type:

PipelineOutput

Returns:

The structured pipeline output.