motosan-agent-workflow 0.3.0

Pure Python implementation of the motosan-agent-workflow SDK

motosan-agent-workflow (Python)

A general-purpose DAG-based agent workflow engine for orchestrating LLM agents.

Installation

pip install motosan-agent-workflow

# Optional YAML support:
pip install "motosan-agent-workflow[yaml]"

Quick Start

import asyncio
from motosan_agent_workflow import (
    Workflow, Node, Runtime, LlmClient, LlmResponse,
)


class MyLlm(LlmClient):
    async def call(self, system_prompt: str, input_data: dict, tools: list[str]) -> LlmResponse:
        return LlmResponse(content={"answer": "..."}, input_tokens=100, output_tokens=50)


async def main():
    workflow = (
        Workflow.builder("my-pipeline")
        .node(Node.agent("researcher", system_prompt="Find key facts about the topic."))
        .node(Node.agent("writer", system_prompt="Write a report.", input_from=["researcher"]))
        .edge("researcher", "writer")
        .build()
    )

    runtime = Runtime(MyLlm())
    result = await runtime.execute(workflow, {"topic": "Python in 2026"})
    print(result.node_output("writer").content)


asyncio.run(main())

YAML Workflow

from motosan_agent_workflow import load_workflow, load_workflow_from_str, Format

# From file
workflow = load_workflow("my-workflow.yaml")

# From string
workflow = load_workflow_from_str(yaml_str, Format.YAML)

Node Types

Type	Description	LLM
Node.agent()	LLM-powered agent	Yes
Node.human()	Human-in-the-loop gate	No
Node.transform()	Pure function transform	No
Node.condition()	Conditional branching	No
Node.loop_node()	Iterative loop	Depends
Node.sub_workflow()	Nested sub-workflow	Depends

Features

• Parallel execution — Nodes in the same DAG layer run concurrently via asyncio.gather
• Output schema validation — JSON Schema with auto self-correction
• Retry policies — Per-node exponential backoff with Skip/Abort/Fallback modes
• Human-in-the-loop — Gates with timeout and default action
• Event streaming — WorkflowEvent via async callback
• Token tracking — Per-node and total, with budget enforcement
• Cost estimation — Pre-execution cost estimates with configurable pricing
• YAML/JSON loader — load_workflow() and load_workflow_from_str()
• 8 built-in templates — code-review, research, brainstorm, report, playlist, content, architecture-review, sprint-planning

Requirements

• Python >= 3.10
• Optional: pyyaml >= 6.0 for YAML support

Publishing

git tag python-v0.3.0
git push origin python-v0.3.0
# → triggers publish-python.yml → PyPI