Company
A frontier lab
that shows its work.
Cogniphic Superintelligence builds intelligence at every layer, from the GPU that runs a model to the app you use it in. We publish what we measure, label what we illustrate, and write our decisions down before we build them.
What we believe
Four rules
we build by.
They decide what ships and what this site is allowed to say.
- Show the work
A system people rely on should say what it read, what it ran, what it cost and what it changed. Ours do, and so does this site.
- Measure, then claim
A number goes on a page only after it came out of a run we can show. The rest stays unsaid until it does.
- Label what is illustrative
Where a figure is sample data or an estimate, it says so beside it, every time it appears.
- Write decisions down
Architectural choices get a record before they get code: what we chose, what we turned down, and why.
The name
Cogniphic
[cog·ni·phic] adjective
Drawn to intelligence; finding understanding beautiful.
Of a system: reasoning in the open, so that anyone can follow it, check it and build on it. The mark is a cube built in steps: solid where the work is done, still stepping into place where it is being built.
Divisions
Seven divisions,
one stack.
Each division ships on its own and makes the others better.
So far
What we have
shipped, in order.
From the repository’s history and its decision records. 169 commits so far.
- The core
Platform decisions, a Rust core beside the desktop shell, safe IPC, the sandbox, the context engine and the verifier.
- The first study
The P12 benchmark: 312 live runs against hidden tests, cost per accepted task down 49%.
- Plugins and Studio
Remote MCP connectors behind approvals, and image and video generation in one panel.
- Cloud
One sandbox account behind a gateway, for runs that need more than a Mac.
- Games
A native Game Studio: describe a game, play it, keep every version.
- Authorised testing
A pentest module with fixed tools that only touch targets you confirm.
Products
Two products,
one company.
Forge
Your whole AI team in one Mac app. Every frontier model, every change tested.
Afterburn
Measure every inference endpoint you run, and route each request on the evidence.