The Agent Toolkit: CrewAI, ADK, LangGraph, OpenAI on Glean

Day 4 Lab

1. Pick one framework your team uses: CrewAI, Google ADK, LangGraph, or the OpenAI Agents SDK.
2. Install Glean's typed client and confirm a bare search call works (you did this in Week 7).
3. Wrap that search call as a tool the framework can call, with a clear name and docstring.
4. Register the tool with one agent and have it answer a question that requires company context.

Working Example: A Glean Search Tool for a Framework Agent

Glean describes building against its platform with frameworks such as CrewAI, Google ADK, LangGraph, and the OpenAI Agents SDK. The common core is small: wrap a typed-client search call as a tool, then let the framework orchestrate. Here the tool is framework-neutral — the same function plugs into any of them. Verified against developer_platform.

import os
from glean import Glean

# One client, authenticated once from the environment — never hard-code tokens.
_client = Glean(
    instance=os.environ["GLEAN_INSTANCE"],
    api_token=os.environ["GLEAN_API_TOKEN"],
)


def glean_search(query: str, top_k: int = 5) -> list[dict]:
    """Search the enterprise graph and return titles + URLs.

    Wrap this as a tool in CrewAI / Google ADK / LangGraph / OpenAI Agents SDK.
    Retrieval runs as the token's user, so results stay permissions-aware.
    """
    resp = _client.client.search.query(query=query, page_size=top_k)
    return [{"title": r.title, "url": r.url} for r in resp.results]


# CrewAI:    Tool(name="glean_search", func=glean_search, description=...)
# OpenAI:    @function_tool def glean_search(...) -> ...
# LangGraph: bind glean_search into a ToolNode
# Google ADK: register glean_search as a FunctionTool

The MAP lesson: whatever the framework, the trust boundary is the same — the tool carries a token that resolves to a user, so the agent can only ever retrieve what that user is allowed to see. Map that once and it holds across every framework. Verified against developer_platform.