First Code (0 → 1)

Goal: write a runnable MASFactory workflow in minutes and understand the essential APIs: RootGraph / Node / Edge / build / invoke.

---

0) Two concepts to understand first

1. MASFactory uses Python dict as the message carrier. Inputs, node outputs, and edge payloads are all structured as dict. The smallest unit of message exchange is a single field in that dict.

• Horizontal message passing: nodes exchange payloads through Edge inside the same graph. Edge.keys defines the field contract and routing rule (“which fields flow across this hop”).
• Vertical message passing: a node exchanges state with its containing graph through attributes (node attributes). pull_keys / push_keys decide which fields are read from / written back to attributes.
• For details, see: Message Passing.

2. RootGraph.invoke(...) returns a tuple: (output_dict, attributes_dict). The first item is the final output of horizontal passing; the second item is a snapshot of the graph attributes (vertical state).

Diagram: Horizontal (Edge) + Vertical (attributes)

---

1) Your first Agent workflow

We build a simple QA workflow with two agents: one for analysis and one for answering.

Diagram: analyze → answer

from masfactory import RootGraph, Agent, OpenAIModel, NodeTemplate, HistoryMemory, Shared, Factory

model = OpenAIModel(
    api_key="YOUR_API_KEY",
    base_url="YOUR_BASE_URL",
    model_name="gpt-4o-mini",
)

BaseAgent = NodeTemplate(
    Agent,
    model=Shared(model),
    memories=[Factory(lambda: HistoryMemory(top_k=12))],
)

g = RootGraph(
    name="qa_two_stage",
    nodes=[
        ("analyze", BaseAgent(instructions="You are a problem analysis expert.", prompt_template="User question: {query}")),
        ("answer", BaseAgent(instructions="You are a solution expert. Provide the final answer based on the analysis.", prompt_template="Question: {query}\nAnalysis: {analysis}")),
    ],
    edges=[
        ("entry", "analyze", {"query": "User question"}),
        ("analyze", "answer", {"query": "Original question", "analysis": "Analysis result"}),
        ("answer", "exit", {"answer": "Final answer"}),
    ],
)

g.build()
out, _attrs = g.invoke({"query": "I want to learn Python. Where should I start?"})
print(out["answer"])

Read model config from environment variables

MASFactory does not auto-load environment variables. You can wire them in yourself:

import os
from masfactory import OpenAIModel

model = OpenAIModel(
    api_key=os.getenv("OPENAI_API_KEY", ""),
    base_url=os.getenv("OPENAI_BASE_URL") or None,
    model_name=os.getenv("OPENAI_MODEL_NAME", "gpt-4o-mini"),
)

Why doesn’t an Agent read attributes by default?

Agent.pull_keys / Agent.push_keys defaults to {}. With this setting, an agent does not read from or write back to graph attributes.
If you want an agent to read from attributes, explicitly set pull_keys to the required fields. If you set pull_keys=None, it will inherit all attributes.

Want reusable capability packs?

If you want to reuse Anthropic-style SKILL.md capability packages across agents, see: Skills.

---

3) What to read next?

• For a systematic understanding of declarative orchestration: read the Development Guide and Examples.
• For runnable workflows: continue with the Examples section, or check applications/*/README.md in this repository.