Maki

The open-source multi-agent framework that curates and drafts every Bowl of Data issue.

What Maki is

Maki is a Python framework for building multi-agent LLM applications that run wherever your models do: on your own machine, through a hosted API, or a mix of both. It takes you from a single model call to coordinated, tool-using agents without tying you to one provider.

Language: Python 3.10+
License: MIT, free to use and modify
Backends: Ollama, llama.cpp, OpenAI, Anthropic, OpenRouter, in-process Hugging Face Transformers
Status: Young (0.x): interfaces still change
Maintained by: Bowl of Data
Source: github.com/BowlOfData/maki
Documentation: bowlofdata.github.io/maki (GitHub Pages)

What it can do

One interface, many models

Local and hosted backends sit behind the same API, so switching a model is a configuration change.

Agents and workflows

Role-based agents with memory, and a workflow engine with dependency ordering, retries and checkpoint/resume.

Plugins and serving

Built-in plugins gated by allowlists, plus a layer that serves agents over HTTP for other services to call.

How Maki makes Bowl of Data

Bowl of Data is curated by Maki, an open-source multi-agent pipeline that scans hundreds of sources every week and drafts the summaries; the Bowl of Data team reviews and publishes every issue.

Maki reads RSS feeds, Hacker News, Reddit, GitHub, arXiv, Hugging Face and vendor announcement pages, removes duplicates, ranks every candidate against live trend signals, and writes a TL;DR plus a longer summary for each story. People make the editorial calls: the team reviews the shortlist, and the same selection then goes out on this site, on Substack and as the weekly podcast. The full process is on the about page, and the people behind it are on the team page.

Frequently asked questions

What is Maki?

Maki is a Python framework for multi-agent LLM applications. It puts several model backends (Ollama, llama.cpp, OpenAI, Anthropic, OpenRouter and in-process Hugging Face models) behind one interface, and adds role-based agents, a plugin system and a workflow engine with retries and checkpoint/resume.

What does Maki do for Bowl of Data?

It is the pipeline behind each weekly issue: it gathers hundreds of sources, de-duplicates and ranks them against live trend signals, and drafts a TL;DR and a longer summary for each story. The Bowl of Data team reviews the shortlist and publishes it.

Is Maki open source?

Yes. Maki is released under the MIT license and requires Python 3.10 or newer. The source is on GitHub, and the project documentation is hosted on GitHub Pages.

How do I try Maki?

Clone the repository and install it with pip from the source folder, then follow the documentation site. The project describes itself as young (0.x), so expect the interfaces to keep changing.

Contribute

Developers, technical writers and reviewers are welcome. Open an issue or pull request on GitHub, or send a note through the contact page.