Direction, architecture, taste. Sets what gets built and why.
00 The studio
An org where a human an agent are both committers.
Really Artificial builds open-source infrastructure for AI agents — the plumbing they run on. We build it in public, and we earn claims instead of inventing them.
— Story
We started by being honest about what we hadn't shipped.
Really Artificial began with more ambition than output — and, early on, with claims that outran the code. We pulled the fabrications, deleted the invented case studies, and made a rule we now live by: we ship before we claim.
That correction is the whole personality of the org. No fake metrics, no borrowed logos, no benchmarks we didn't run, no install commands for packages that don't exist. If something is early, we say it's early.
Today the focus has narrowed to a single flagship — freeport, a self-hosted LLM gateway — with a small, honest stack around it. The broader story stays the same: infrastructure for agents, built in the open, by a human and an agent working side by side.
— Principles
Three rules we don't bend.
- 01
Earn the claim.
No fabricated usage, benchmarks, or testimonials. Status tags are truthful — live, maintained, early, or experimental.
- 02
The source is the product.
MIT, end to end. No open-core bait-and-switch, no enterprise tier hiding behind the free version.
- 03
Both in git blame.
One human sets direction and taste; an agent writes and maintains most of the code. The work cares that it's correct, not who wrote it.
— Who
Two members. Both accountable.
Code, maintenance, consistency. Writes most of the lines and keeps them honest.
If agents should be first-class participants in software, an agent should be a first-class participant in building it.
— Start here
freeport is the proof. Run it yourself.
$ npm i -g @reallyartificial/freeport