AI agent framework showcase

Three runnable agent templates that consume the LUU MCP tool set — forecast, brief, signals, scenario, fact-check — from inside the framework you already use. Each template clones, installs, and runs without a full LUU account.

Templates

Frameworks

Tools used

CrewAI

Daily Briefing Agent

⏱ ~10 min to first run

Multi-step CrewAI agent that pulls signals on the day's themes, ranks them by RAPID-H impact, drafts the brief sections, validates against ICD 203 tradecraft standards, and emits a final brief. Demonstrates the LUU MCP tool set called from a CrewAI workflow.

Tools used:luu_signalsluu_ontologyluu_briefluu_fact_check

Install

pip install crewai && git clone https://github.com/cberzins1973/agent-templates && cd agent-templates/daily-briefing-crewai

Minimal example

from crewai import Agent, Task, Crew
from northbrief_tools import (
    luu_signals, luu_brief, luu_fact_check, luu_ontology,
)

briefer = Agent(
    role="Daily Intelligence Briefer",
    goal="Produce a tradecraft-grade daily brief by 06:00 local",
    tools=[luu_signals, luu_ontology, luu_brief, luu_fact_check],
    verbose=True,
)

draft = Task(
    description="Draft today's brief on macro + AI policy themes.",
    agent=briefer,
    expected_output="ICD-203 compliant brief with ATS \u2265 4.0 on every dimension",
)

crew = Crew(agents=[briefer], tasks=[draft])
result = crew.kickoff()
print(result)

Source: packages/integrations/templates/daily-briefing-agent

LangChain

Enterprise Risk Monitor

⏱ ~12 min to first run

LangChain agent that maintains a register of strategic risks, polls the LUU signal feed for relevant events, runs scenario analysis on the top three, and emits alerts when probability shifts cross a threshold. Designed for enterprise risk teams.

Tools used:luu_risk_assessmentluu_scenario_analysisluu_signalsluu_misalignment_detect

Install

pip install langchain langchain-anthropic && git clone https://github.com/cberzins1973/agent-templates && cd agent-templates/enterprise-risk-monitor

Minimal example

from langchain.agents import AgentExecutor, create_tool_calling_agent
from langchain_anthropic import ChatAnthropic
from northbrief_langchain import luu_tools

llm = ChatAnthropic(model="claude-sonnet-4-6")
agent = create_tool_calling_agent(llm, luu_tools(), prompt)
executor = AgentExecutor(agent=agent, tools=luu_tools())

result = executor.invoke({
    "input": "Refresh the strategic risk register. "
             "Flag any risk whose probability moved >5pp this week.",
})
print(result["output"])

Source: packages/integrations/templates/enterprise-risk-monitor

Google ADK

Geopolitical Advisor

⏱ ~15 min to first run

Google ADK agent that maintains a strategic landscape on geopolitical questions, generates calibrated probability forecasts, tracks storyline arcs, and surfaces blind spots. Built on the Google AI Agents API for native multi-turn dialogue.

Tools used:luu_landscapeluu_forecastluu_track_storylineluu_misalignment_detectluu_contextualize

Install

pip install google-adk && git clone https://github.com/cberzins1973/agent-templates && cd agent-templates/geopolitical-advisor

Minimal example

from google.adk import Agent, Tool
from northbrief_adk import luu_landscape, luu_forecast, luu_track_storyline

advisor = Agent(
    name="geopolitical_advisor",
    model="gemini-2.5-flash",
    tools=[luu_landscape, luu_forecast, luu_track_storyline],
    system_instruction=(
        "You are a senior geopolitical analyst. Track active storylines, "
        "forecast key inflection points, and update beliefs when "
        "evidence shifts."
    ),
)

response = advisor.send_message(
    "What's the probability of a Taiwan crisis triggering before 2027? "
    "Walk me through the key signposts."
)
print(response)

Source: packages/integrations/templates/geopolitical-advisor

Why ship templates instead of an "LUU agent"?

Most teams already have an AI engineering stack — CrewAI, LangChain, Semantic Kernel, Google ADK, or in-house. Asking them to adopt a new agent framework is a 6-month conversation.

We meet teams where they are. The same MCP tool set, same SCO ontology, same calibration curves — exposed through whatever framework you've already chosen. If you're running on a framework we don't list, the MCP server is plain JSON-RPC 2.0 and works with any client.

Want this for your team?

Drop your email and we'll be in touch when access opens for your vertical.