Build a research assistant with Maki

The course

You build Scout, a small research assistant, one step at a time. It runs on a local model through Ollama, so you need no API key. Each chapter ends with something that works.

1. 1 What Maki Is and What You Will Build Start here: what Maki is, when it is the right tool, and the small research assistant, Scout, that you will build on your own machine.
2. 2 Install Maki and Run Your First Agent Create a Maki agent in a short Python program, run it on a local model with Ollama, llama.cpp or vLLM, and understand each part of it.
3. 3 Give Your Agent a Tool Let a Maki agent call a tool: load the web_search plugin, ask for Hacker News stories, and learn how the agent decides when to use it.
4. 4 Build a Two-Step Pipeline Combine plain Python and two Maki agents into a pipeline that finds a story, reads the page and writes a short brief, and learn where it breaks.

Guides

• Build a Multi-Agent LLM Pipeline in Python A working multi-agent LLM pipeline in Python: plain code fetches, one agent reads, another writes. Complete code, runs locally with no API key.
• Run LLM Agents Locally with Ollama in Python Run an LLM agent in Python on your own computer with Ollama and Maki: install, pull a small model, and get an answer with no API key.

What you need

Python

3.10 or newer

Install

pip install maki-framework

Model

Ollama and qwen3.5:4b (3.4 GB), pulled before you start

Last tested

2026-11-12, with the latest Maki release