Build a collaborative research agent

Build an AgentApp that can search and fetch public web sources through multiple bounded rounds of model-directed tool use. It preserves conversation messages, recovers from connector failures, and always ends its tool loop.

The finished project uses the complete public AgentSession surface:

• agent.responses.create for model requests

• agent.connectors.tools for runtime-provided schemas

• agent.connectors.call for function calls

It uses only web_search and web_fetch. Neither requires an external account.

Create the project

$ mkdir research-agent
$ cd research-agent
$ mkdir research_agent
$ touch README.md research_agent/__init__.py

Create:

research-agent/
├── .gitignore
├── README.md
├── pyproject.toml
└── research_agent/
    ├── __init__.py
    └── agent_app.py

Add .gitignore:

.venv/
*.fab
__pycache__/

Add README.md:

# Research Agent

A bounded Flower AgentApp that researches public web sources with `web_search`
and `web_fetch`.

Configure the project

The project configuration has three jobs:

• pin the Flower version used by the AgentApp

• provide a default agent.input value for each run

• tell Flower where to load the AgentApp object

Create pyproject.toml:

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "research-agent"
version = "0.1.0"
description = "A bounded public-web research AgentApp"
license = "Apache-2.0"
requires-python = ">=3.11"
dependencies = ["flwr==1.35.0"]

[tool.hatch.build.targets.wheel]
packages = ["research_agent"]

[tool.flwr.app]
publisher = "local"
display-name = "Research Agent"
flwr-version-target = "1.35.0"
fab-include = ["research_agent/**/*.py"]

[tool.flwr.app.config.agent]
input = "Find two public sources that explain federated AI and compare them."

[tool.flwr.app.components]
agentapp = "research_agent.agent_app:app"

Implement the AgentApp

Build research_agent/agent_app.py one section at a time. Add the following snippets in order.

Define the app and its limits

Every AgentApp entry point receives two objects:

• AgentSession provides model responses and connector calls

• Context provides run configuration and state shared by the run series

Keep the model, connector set, and tool-turn limit near the top of the file. The turn limit is a safety boundary: the model can request more work, but it cannot keep the app in an unbounded tool loop.

from __future__ import annotations

import json
from typing import Any

from flwr.agentapp import AgentApp, AgentSession
from flwr.app import Context

MODEL = "openai/gpt-5.6-sol"
TOOL_REFS = ("web_search", "web_fetch")
MAX_TOOL_TURNS = 3

app = AgentApp()

Rebuild conversation input

Each chat message starts a new run. Flower keeps the runs together in a run series, but the model sees only the input passed to agent.responses.create. To support follow-up questions, the AgentApp must replay the stored user and assistant messages.

Flower stores run-series state in Context.state. Conversation items live in a ConfigRecord named items, which you can read through context.state.config_records. The state also contains tool activity, so load only items whose type is message and normalize their content to plain text:

def message_text(content: Any) -> str:
    """Normalize a stored Responses message to plain text."""
    if isinstance(content, str):
        return content
    if isinstance(content, list):
        parts = []
        for part in content:
            if not isinstance(part, dict):
                raise TypeError("Message content parts must be objects")
            value = part.get("text", part.get("refusal"))
            if not isinstance(value, str):
                raise TypeError("Message content parts must contain text or refusal")
            parts.append(value)
        return "\n".join(parts)
    raise TypeError("Message content must be text or a list of content parts")


def conversation_messages(context: Context) -> list[dict[str, Any]]:
    """Replay only user and assistant messages from the run series."""
    messages: list[dict[str, Any]] = []
    items_record = context.state.config_records.get("items")
    items = items_record.get("json", []) if items_record is not None else []
    for item_json in items:
        item = json.loads(item_json)
        if item.get("type") != "message":
            continue
        messages.append(
            {
                "type": "message",
                "role": item["role"],
                "content": message_text(item["content"]),
            }
        )
    return messages

message_text handles plain strings and Responses-style text or refusal parts. It raises an error for an unexpected shape instead of silently sending incomplete history to the model.

Keep planning responses private

agent.responses.create retains response items in Context so that an assistant answer can become part of the conversation. The first responses in this app are different: they are private planning turns in which the model can request tools.

Use the same config_records view to snapshot the stored items before a planning request and restore them afterward. The tool calls still remain in the local input_items list for the current run, but draft model output does not become conversation history:

def private_response(
    agent: AgentSession,
    context: Context,
    request: dict[str, Any],
) -> dict[str, Any]:
    """Make a planning request without retaining its draft model output."""
    items_record = context.state.config_records.get("items")
    previous_items = (
        list(items_record.get("json", ())) if items_record is not None else []
    )
    try:
        return agent.responses.create(request)
    finally:
        if items_record is not None:
            items_record["json"] = previous_items
        elif "items" in context.state.config_records:
            del context.state.config_records["items"]

Let the model recover from connector failures

A connector can fail after the model has requested it, and the model can return malformed connector arguments. The next model turn still needs an output for that call ID. Convert the exception into a function_call_output item so the model can explain the limitation or finish with the evidence it already has:

def connector_error_output(
    tool_call: dict[str, Any], exc: Exception
) -> dict[str, Any]:
    """Return an error item the model can handle in its next turn."""
    return {
        "type": "function_call_output",
        "call_id": tool_call["call_id"],
        "output": json.dumps({"error": str(exc)}),
    }

Orchestrate the tool loop

The main function now connects these pieces. It has four phases:

1. Validate agent.input and rebuild the conversation messages

2. Ask Flower for the web_search and web_fetch tool schemas

3. Execute up to MAX_TOOL_TURNS rounds of model-requested function calls

4. Make one final model request without tools and stream the answer

Add the entry point:

@app.main()
def main(agent: AgentSession, context: Context) -> None:
    """Research the configured prompt with a bounded connector loop."""
    prompt = context.run_config.get("agent.input")
    if not isinstance(prompt, str) or not prompt.strip():
        raise ValueError("agent.input must be a non-empty string")

    input_items = conversation_messages(context)
    if not any(
        item["role"] == "user" and item["content"].strip() == prompt.strip()
        for item in input_items
    ):
        input_items.append(
            {"type": "message", "role": "user", "content": prompt.strip()}
        )

    tools = agent.connectors.tools(TOOL_REFS)
    allowed_tool_names = {
        tool["name"] for tool in tools if isinstance(tool.get("name"), str)
    }

    for _ in range(MAX_TOOL_TURNS):
        response = private_response(
            agent,
            context,
            {
                "model": MODEL,
                "input": input_items,
                "instructions": (
                    "Research the user's question using public sources when useful. "
                    "Request all independent tool calls for a turn together."
                ),
                "tools": tools,
                "tool_choice": "auto",
                "stream": False,
            },
        )
        response_output = [
            dict(item)
            for item in response.get("output", [])
            if isinstance(item, dict)
        ]
        tool_calls = [
            item for item in response_output if item.get("type") == "function_call"
        ]
        if not tool_calls:
            break

        function_outputs = []
        for tool_call in tool_calls:
            if tool_call.get("name") not in allowed_tool_names:
                function_outputs.append(
                    connector_error_output(
                        tool_call,
                        RuntimeError(
                            f"Tool {tool_call.get('name')!r} was not exposed"
                        ),
                    )
                )
                continue
            try:
                arguments = tool_call.get("arguments")
                if isinstance(arguments, str):
                    arguments = json.loads(arguments)
                if not isinstance(arguments, dict):
                    raise ValueError("Tool call arguments must be a JSON object")
                function_outputs.append(agent.connectors.call(tool_call))
            except (RuntimeError, ValueError) as exc:
                function_outputs.append(connector_error_output(tool_call, exc))

        input_items.extend(response_output)
        input_items.extend(function_outputs)

    agent.responses.create(
        {
            "model": MODEL,
            "input": input_items,
            "instructions": (
                "Answer the user's question from the available evidence. "
                "Mention any failed source access and do not invent results."
            ),
            "stream": True,
        }
    )

The runtime records the current agent.input as a user message before calling the app. The duplicate check prevents the same prompt from being appended again. Within the loop, the app keeps the complete model output, including reasoning items and every requested call, next to the connector output. This gives the next model turn a complete sequence even when a connector fails or the model requests a tool that was not exposed. The allowed names come from the tool schemas rather than TOOL_REFS because one connector reference can expose several tools. The final request omits tools, which forces the app to finish with one answer instead of starting another connector round.

Note

The connector activity itself is still recorded for run inspection. The app replays only message items on the next run, so connector events and orphaned function outputs are not treated as conversation messages.

Copy the complete file

If you prefer to start from the finished version, expand the block below and copy it into research_agent/agent_app.py.

Complete research_agent/agent_app.py

from __future__ import annotations

import json
from typing import Any

from flwr.agentapp import AgentApp, AgentSession
from flwr.app import Context

MODEL = "openai/gpt-5.6-sol"
TOOL_REFS = ("web_search", "web_fetch")
MAX_TOOL_TURNS = 3

app = AgentApp()


def message_text(content: Any) -> str:
    """Normalize a stored Responses message to plain text."""
    if isinstance(content, str):
        return content
    if isinstance(content, list):
        parts = []
        for part in content:
            if not isinstance(part, dict):
                raise TypeError("Message content parts must be objects")
            value = part.get("text", part.get("refusal"))
            if not isinstance(value, str):
                raise TypeError("Message content parts must contain text or refusal")
            parts.append(value)
        return "\n".join(parts)
    raise TypeError("Message content must be text or a list of content parts")


def conversation_messages(context: Context) -> list[dict[str, Any]]:
    """Replay only user and assistant messages from the run series."""
    messages: list[dict[str, Any]] = []
    items_record = context.state.config_records.get("items")
    items = items_record.get("json", []) if items_record is not None else []
    for item_json in items:
        item = json.loads(item_json)
        if item.get("type") != "message":
            continue
        messages.append(
            {
                "type": "message",
                "role": item["role"],
                "content": message_text(item["content"]),
            }
        )
    return messages


def private_response(
    agent: AgentSession,
    context: Context,
    request: dict[str, Any],
) -> dict[str, Any]:
    """Make a planning request without retaining its draft model output."""
    items_record = context.state.config_records.get("items")
    previous_items = (
        list(items_record.get("json", ())) if items_record is not None else []
    )
    try:
        return agent.responses.create(request)
    finally:
        if items_record is not None:
            items_record["json"] = previous_items
        elif "items" in context.state.config_records:
            del context.state.config_records["items"]


def connector_error_output(
    tool_call: dict[str, Any], exc: Exception
) -> dict[str, Any]:
    """Return an error item the model can handle in its next turn."""
    return {
        "type": "function_call_output",
        "call_id": tool_call["call_id"],
        "output": json.dumps({"error": str(exc)}),
    }


@app.main()
def main(agent: AgentSession, context: Context) -> None:
    """Research the configured prompt with a bounded connector loop."""
    prompt = context.run_config.get("agent.input")
    if not isinstance(prompt, str) or not prompt.strip():
        raise ValueError("agent.input must be a non-empty string")

    input_items = conversation_messages(context)
    if not any(
        item["role"] == "user" and item["content"].strip() == prompt.strip()
        for item in input_items
    ):
        input_items.append(
            {"type": "message", "role": "user", "content": prompt.strip()}
        )

    tools = agent.connectors.tools(TOOL_REFS)
    allowed_tool_names = {
        tool["name"] for tool in tools if isinstance(tool.get("name"), str)
    }

    for _ in range(MAX_TOOL_TURNS):
        response = private_response(
            agent,
            context,
            {
                "model": MODEL,
                "input": input_items,
                "instructions": (
                    "Research the user's question using public sources when useful. "
                    "Request all independent tool calls for a turn together."
                ),
                "tools": tools,
                "tool_choice": "auto",
                "stream": False,
            },
        )
        response_output = [
            dict(item)
            for item in response.get("output", [])
            if isinstance(item, dict)
        ]
        tool_calls = [
            item for item in response_output if item.get("type") == "function_call"
        ]
        if not tool_calls:
            break

        function_outputs = []
        for tool_call in tool_calls:
            if tool_call.get("name") not in allowed_tool_names:
                function_outputs.append(
                    connector_error_output(
                        tool_call,
                        RuntimeError(
                            f"Tool {tool_call.get('name')!r} was not exposed"
                        ),
                    )
                )
                continue
            try:
                arguments = tool_call.get("arguments")
                if isinstance(arguments, str):
                    arguments = json.loads(arguments)
                if not isinstance(arguments, dict):
                    raise ValueError("Tool call arguments must be a JSON object")
                function_outputs.append(agent.connectors.call(tool_call))
            except (RuntimeError, ValueError) as exc:
                function_outputs.append(connector_error_output(tool_call, exc))

        input_items.extend(response_output)
        input_items.extend(function_outputs)

    agent.responses.create(
        {
            "model": MODEL,
            "input": input_items,
            "instructions": (
                "Answer the user's question from the available evidence. "
                "Mention any failed source access and do not invent results."
            ),
            "stream": True,
        }
    )

Build and run

$ uv sync
$ uv run flwr build
$ uv run flwr login supergrid
$ uv run flwr run . supergrid --stream

Override the research prompt:

$ uv run flwr run . supergrid \
    --run-config 'agent.input="Compare two recent public explanations of federated AI."' \
    --stream

Success checkpoint

The run finishes with one streamed answer. In SuperGrid run activity, you can see zero or more search/fetch calls and any connector failure that the final answer had to handle.

Adapt it safely

• Keep TOOL_REFS limited to the capabilities the task needs

• Keep a finite tool-turn limit even when you change models

• Validate every required run-config value before making a model call

• Never put credentials in prompts or connector arguments

• Use Connect accounts before adding an account connector, and remember that those runs are personal-workspace-only

• Follow Create automations before exposing start_automation for explicit future or recurring requests