Example Agents

PyFlue is inspired by the agent examples shown by the Flue framework. This page shows the same product-oriented agent categories in Python:

• issue triage
• data analyst
• coding agent
• support agent

The examples live under:

examples/agents/

Coding Agent

This is the Python version of the "build your own coding agent" pattern. It creates an agent with shell access, prepares a repository workspace, and sends a coding prompt to the session.

from pydantic import BaseModel
from pyflue import init


class CodingResult(BaseModel):
    summary: str
    files_changed: list[str] = []
    tests_run: list[str] = []


async def run_coding_agent(repo: str, prompt: str) -> CodingResult:
    agent = await init(
        model="openai:gpt-5.5",
        sandbox="virtual",
        allow_write=True,
        allow_shell=True,
        allowed_commands=["git", "mkdir", "python"],
    )
    session = await agent.session("code")

    await session.shell("mkdir -p workspace")
    await session.shell(f"git clone {repo} workspace/project")
    await session.shell("python -m pip install -e .", timeout=600)

    return await session.prompt(prompt, result=CodingResult)

Full file:

examples/agents/coding_agent.py

Issue Triage

The issue triage agent accepts issue data, uses a Markdown skill, and returns a typed Pydantic result.

from pydantic import BaseModel
from pyflue import init


class TriageResult(BaseModel):
    severity: str
    summary: str
    suggested_labels: list[str] = []
    fix_recommended: bool = False


async def triage_issue(issue_number: int, title: str, body: str) -> TriageResult:
    agent = await init(model="openai:gpt-5.5")
    session = await agent.session(f"issue-{issue_number}")
    return await session.skill(
        "triage_issue",
        args={"issue_number": issue_number, "title": title, "body": body},
        result=TriageResult,
    )

Full file:

examples/agents/issue_triage.py

Data Analyst

The data analyst agent writes data into the sandbox, uses shell tools to inspect it, and asks the model for structured analysis.

from pydantic import BaseModel
from pyflue import init


class AnalysisResult(BaseModel):
    summary: str
    metrics: dict[str, float]
    recommendations: list[str]


async def analyze_csv(csv_text: str, question: str) -> AnalysisResult:
    agent = await init(
        model="openai:gpt-5.5",
        python_backend="monty",
        allow_write=True,
    )
    session = await agent.session("data-analysis")
    await session.write_file("data/input.csv", csv_text)
    preview = await session.run_python(
        "len(rows)",
        inputs={"rows": csv_text.splitlines()},
    )
    return await session.prompt(
        f"Analyze data/input.csv for this question: {question}\n\nRows: {preview.result}",
        result=AnalysisResult,
    )

Full file:

examples/agents/data_analyst.py

Support Agent

The support agent writes knowledge base articles into the sandbox, searches them with grep, and returns a typed answer.

from pydantic import BaseModel
from pyflue import init


class SupportAnswer(BaseModel):
    answer: str
    cited_articles: list[str]
    confidence: str


async def answer_support_question(question: str, articles: dict[str, str]) -> SupportAnswer:
    agent = await init(model="openai:gpt-5.5", allow_write=True, allow_shell=True)
    session = await agent.session("support")

    for name, content in articles.items():
        await session.write_file(f"knowledge/{name}.md", content)

    matches = await session.shell(f"grep -Rni {question.split()[0]!r} knowledge || true")
    return await session.prompt(
        f"Answer this customer question using knowledge/*.md.\n\nQuestion: {question}\n\nSearch matches:\n{matches['stdout']}",
        result=SupportAnswer,
    )

Full file:

examples/agents/support_agent.py

Running Examples

Set a provider key for model-backed examples:

export OPENAI_API_KEY=...
python examples/agents/coding_agent.py

The examples are meant to be copied into your own project and adjusted for your model provider, sandbox policy, and deployment target.