@agentic_function Calling Sub-Functions in a Fixed Order#

Optionally calls the LLM, then invokes multiple sub-functions in the order hard-coded in the source.

Use Cases#

• Research flow: survey → find gap → generate ideas
• Paper flow: write draft → review → revise
• Data flow: collect → clean → analyze
• Any multi-step task with a fixed step order

Design Points#

• Use the @agentic_function decorator
• Call multiple sub-@agentic_functions in a fixed order
• exec() is optional: don't call it (pure chaining), or call it multiple times (each call creates an exec child node)
• Data is passed between sub-functions through Python variables
• A single function can call exec() multiple times, and can call any number of other @agentic_functions

Example: No exec, Pure Chaining#

@agentic_function
def research_pipeline(task: str, runtime: Runtime) -> dict:
    """Run the full research flow: survey → find gap → generate ideas.

    Args:
        task: Research topic.
        runtime: LLM runtime instance.

    Returns:
        A result dict containing survey, gaps, and ideas.
    """
    survey = survey_topic(topic=task, runtime=runtime)
    gaps = identify_gaps(survey=survey, runtime=runtime)
    ideas = generate_ideas(gaps=gaps, runtime=runtime)

    return {"survey": survey, "gaps": gaps, "ideas": ideas}

Example: Calling exec Once to Summarize#

@agentic_function
def research_pipeline(task: str, runtime: Runtime) -> str:
    """Run the full research flow and summarize the result.

    Args:
        task: Research topic.
        runtime: LLM runtime instance.

    Returns:
        The consolidated research summary.
    """
    survey = survey_topic(topic=task, runtime=runtime)
    gaps = identify_gaps(survey=survey, runtime=runtime)
    ideas = generate_ideas(gaps=gaps, runtime=runtime)

    return runtime.exec(content=[
        {"type": "text", "text": (
            f"Survey:\n{survey}\n\n"
            f"Gaps:\n{gaps}\n\n"
            f"Ideas:\n{ideas}"
        )},
    ])

Context Tree#

research_pipeline
├── survey_topic       ← step 1
├── identify_gaps      ← step 2
└── generate_ideas     ← step 3

Passing Data Between Steps#

Data is passed between sub-functions through Python variables, with no LLM involvement:

survey = survey_topic(topic=task, runtime=runtime)
gaps = identify_gaps(survey=survey, runtime=runtime)

The return value of survey_topic is used directly as the input argument to identify_gaps.

Inserting Python Processing Between Steps#

survey = survey_topic(topic=task, runtime=runtime)

# Insert ordinary Python processing in between
key_points = extract_key_points(survey)
filtered = [p for p in key_points if p["relevance"] > 0.5]

gaps = identify_gaps(survey="\n".join(filtered), runtime=runtime)

Error Handling#

survey = survey_topic(topic=task, runtime=runtime)
if not survey or "error" in survey.lower():
    return {"error": "Survey failed", "survey": survey}

gaps = identify_gaps(survey=survey, runtime=runtime)

Difference from "LLM-Chosen Calls"#