About

The project

Why it exists

Generative AI has become the default reflex. It gets called to filter a message, pull out a date, sort a form, search three hundred documents. In a large share of those cases, a rule, a query or a classic model of a few megabytes does the job, often better.

That reflex has a price few people work out at the moment of the decision: a recurring cost that grows with usage, latency, data leaving your infrastructure, a non-deterministic result that is therefore hard to guarantee, a dependency on a provider and on its deprecations, a regulatory scope you did not choose, and an energy footprint out of all proportion to the alternative.

The most exposed are those with no in-house expert and no consulting budget: junior developers, solo developers, small teams. They copy what they see, and what they see is an ecosystem with an interest in selling them AI.

Overkill exists to make the trade-off available. For each task, every known option from the simplest to the heaviest, with code that runs, orders of magnitude, real limits, and the precise condition that justifies moving up a rung.

What it refuses to be

This is not an anti-AI site. That point is not negotiable. The principle is “the right tool for the job”, not “AI is bad”. Some entries recommend a general-purpose model without hedging, and those are what make the entries recommending the opposite worth believing.

Nor is it a tool directory, an opinion blog, a legal advice service, or a piece of software as a service. No accounts, no payment, no database, no comments.

How it is funded

It is not.

No advertising, no partnerships, no affiliate programme, no sponsored content. Nobody pays to appear on an entry, and nobody can pay to disappear from one.

That has a direct consequence for what you read: where an entry names a tool, it is because the code needs it to run. There is nothing to gain by steering you to the most expensive rung, and nothing to lose by telling you to buy nothing.

It also has a consequence for pace: the catalogue grows at the speed of what can be written and verified properly, not at the speed of a publishing calendar.

Hosting a static site this size costs little. The help page offers services related to the subject; that is the only thing resembling a business model, and it is kept separate from the content.

How it is built

The site is static, with no database and no tracker. The content lives in a public Git repository: each entry is a file, each code snippet is a real file with a test that runs it on every build of the site.

That constraint is not cosmetic. A catalogue whose code does not run is worthless, and the only way to guarantee it is to refuse to publish anything that has not been executed. See the methodology page.

The weight of each page is shown in the footer. A site arguing for frugality that weighed three megabytes would not be credible.

The author

Overkill is written by Florentin Biaou, an AI and data engineer.

To get in touch, the help page gives an address.

Reviewed on September 10, 2026