About
Practical instruments for understanding and building reliable AI systems.
This is a workbench, not a demo reel. Every page is meant to be used, not just read: you put something in, you get an inspectable result out, and you can see exactly which assumptions produced it.
Who this is for
- Architects
- explaining an AI system to a Customer or a team, who need a working instrument on screen instead of a slide.
- Builders
- making a design decision before writing code: what a request costs, what fits in a context window, how a prompt actually behaves.
- Operators
- diagnosing production behavior after the system is live, where a hunch is not a diagnosis.
- Curious technical users
- who learn a system fastest by manipulating it, not by reading a description of it.
What actually constrains the build
These are not aspirations. They are the rules a tool has to pass before it ships:
- Every page must let you do something. Reading about a concept is not the point; using an instrument is.
- An input produces an inspectable result, not generic prose.
- Assumptions are visible and editable, never a hidden default baked into a black box.
- A result is exportable or shareable where that is genuinely useful, not as a default gimmick.
- A tool teaches through use. It should not turn into a textbook chapter you have to read first.
- Every instrument shares one visual language. This reads as one Nixfred laboratory, not a pile of unrelated demos.
- A tool works with sample data out of the box. Its core experience never requires an API key.
- No user input leaves your browser unless the interface explicitly says so.
What this is not
- A general AI chat interface. Every tool here does one legible job, not open ended conversation.
- A news feed, a trend roundup, an essay collection, or a home for speculative dead internet material. This is a workbench, not a magazine.
- A full production agent runtime. These instruments help you design and reason about agents; none of them deploys or operates one on your behalf.
- An account system. No billing and no team collaboration, at least not in this first release.
One commitment sits underneath all of it. A simulated result is never presented as a real model benchmark. If a number on this site is illustrative rather than measured against an actual model, the interface says so, in the interface, not in a footnote.
The catalog is under construction
Every tool in the catalog currently reads coming soon. That is deliberate, not neglect. The shared platform and this landing experience are built and working; each individual tool is its own PRD, reviewed and accepted on its own before it ships, one at a time, under a build control protocol. A tool's card stops reading coming soon only once its own PRD has actually been implemented. Nothing here should be mistaken for a working tool before that happens.
See how the site handles anything you type into it on the privacy page.