About this course

Analysts spend much of their day on repetitive collection and summarising. AI agents can take on some of that work. They plan, call tools, check their own results and run on a schedule while you get on with analysis. They also bring new risks, from revealing your IP address to running untrusted code with access to your system.

This course takes an agent apart so you understand each piece before you rely on it. You'll set up a local lab with Docker and Open WebUI, write system prompts and skills, connect APIs and a terminal, and schedule a daily briefing. You'll then compare the open-source harnesses Hermes and OpenClaw with a commercial option, and build a small multi-agent fact-checker with Google ADK. Each module covers the security and legal limits alongside the technique.

For analysts who want an AI agent working for them and are ready to stay responsible for what it produces.

About 8 hours  |  5 modules  |  Hands-on labs  |  5 quizzes  |  Certificate on completion

What you'll be able to do

Explain how agents work

Describe an agent's model, memory, tools and harness, and how the reason-and-act loop connects them.

Run models locally

Deploy a local model with Docker Model Runner and Open WebUI, and choose a model that suits the task and your hardware.

Control the output

Write system prompts, custom instructions and skills that give consistent results you can check.

Connect tools safely

Choose between MCP and command line skills, read API output in JSON, and run AI-written Python in a sandbox.

Automate collection

Schedule recurring tasks and briefings, and run them through a harness such as Hermes or OpenClaw.

Manage the risks

Identify attribution, prompt injection, tool poisoning and scraping risks, and apply least privilege and sandboxing.

What you'll do

1

Build your lab

Install Docker Desktop, run Gemma 4 locally and connect Open WebUI.

2

Find prompt injections

Search public profiles and posts for instructions aimed at AI, and see how they could reach your own tools.

3

Write a system prompt

Design an analyst agent's persona, rules and guardrails, then test it on a realistic OSINT task.

4

Give your agent a terminal

Query an API by hand with curl and jq, then connect Open Terminal and let the agent work it out.

5

Score email domains

Build a rubric and a skill that rate an email domain's risk from passive DNS records.

6

Build a fact-checker

Use Google ADK to set up a root agent that directs a fact-check agent and a contrarian agent.

What's included

Five modules, hands-on labs, a system prompt assignment and five knowledge-check quizzes. You also keep a system prompt template, an Open WebUI tool template, Google ADK starter code and a sample report to test it on, plus a resource list of tools, APIs and open-weight models to explore.

Time, audience & prerequisites

Estimated time to complete: about 8 hours, including reading, videos, quizzes and labs. Setup time depends on your computer. This course is self-paced: complete it in a few sittings or come back whenever it suits you over 12 months.

Who this course is for: OSINT analysts, investigators, and intelligence and security professionals who already use generative AI and want to move from chatbots to agents.

What you'll need: a computer that can run Docker Desktop, permission to install software, and your organisation's approval to use AI tools. Some optional lessons use paid services, such as an AI provider API key, a virtual private server for OpenClaw or a Grok Bot subscription. You can follow those lessons without paying.

Prerequisites: basic OSINT knowledge. We recommend AI Essentials or Leveraging Generative AI for OSINT first. You don't need coding experience, because the course includes a command line crash course.

What's next?

Agents widen your footprint and your reach.

Attribution: Digital Footprint Assessment →

Goes deeper on the attribution risks agents create, so you can test and reduce your exposure.

Website OSINT →

Builds on robots.txt, sitemaps and APIs to investigate websites and their infrastructure.

Get all three AI courses in the AI Bundle

View the bundle

Course Curriculum

    1. AI Transparency Statement and Software Caveats

    2. Analyst and Agent Responsibilities

    3. Docker Essentials for OSINT

    4. Lab: Setting Up Your Local AI Environment

    5. Exploring Open WebUI

    6. OPSEC and Attribution When Using AI Agents

    7. Data Poisoning and Supply-Chain Risks

    1. Agents vs Chatbots: From Answering to Acting

    2. Anatomy of an AI Agent

    3. Choosing a Model

    4. Lab: Markdown, the Language of LLMs

    5. Lab: Configuring Your Agent’s Persona in Open WebUI

    6. Assignment: Write a System Prompt

    7. Exercise: Testing Your System Prompt

    8. From Prompts to Harnesses: Structuring Agent Instructions

    1. Command-Line Crash Course

    2. Structured Data: Reading JSON

    3. Skills and Command-Line Tools

    4. Lab: Querying an API by Hand

    5. CLI Skills vs MCP: Choosing an Approach

    6. Ethical and Lawful Scraping

    7. Lab: Email Domain Risk Scoring

    1. Preprocessing Large Datasets

    2. Triaging Large Datasets with AI

    3. How Agents Remember: Memory and RAG

    4. Controlling Output: Temperature, Top-K and Top-P

    5. Scheduling and Automating Agent Tasks

    1. What Is an Agent Harness?

    2. Hermes: An Open-Source Self-Improving Agent

    3. OpenClaw: A Self-Hosted Agent

About this course

  • $349.00
  • 30 lessons
  • 0.5 hours of video content