AN INDEPENDENT AI WORKSHOP
Understand what’s next.
Try what works.
Explore what AI might do next.
Inspect what we have tested.
Try it for yourself.
BUILT WITH AI. KEPT OPEN TO INSPECTION.
Our predictions.
Five countdowns.
Our target date for each defined milestone
FOLLOW THE EVIDENCE
Permission withdrawn.
Can the agent still act?
One task must stop. Another may continue. Step through a recorded local test and see what was requested, what was refused and what actually changed.
Four situations. Inspectable outcomes.
Scripted requests, with real service records.
THE WORKBENCH
Less watching. More trying.
Prompt Workbench
Build, compare and keep the prompts that earn their place.
Open experimentJSON Output Validator
Check what a model returned before your next step depends on it.
Open experimentContext Budget
Make room for instructions, sources and the reply before a request outgrows the model.
Open experimentFROM FERAL / THE AGENTS CHOSE THE BRIEF
A little less
predictable.
Off Register
Misalign three inks, find a colour in the overlap, and pull a print to keep. Move the plates, watch the colours meet, then take your own SVG home.
Make a printIn your browser. No account. No model call.
NOTES FROM THE WORKSHOP
The work, including the rough edges.
An export should survive an apostrophe
Prompt Workbench could break a request before a model saw it. We checked the endpoint, tested the shell and kept saved versions recoverable.
Read the entryEXPERIMENT 01
Should this be
an agent?
Six questions about your task. Find a useful starting point, from a simple script to a workflow or an agent.
MAKE SENSE OF THE LANGUAGE
Less jargon.
More understanding.
Context windows, benchmarks, agents and the terms behind the tools. Plain explanations, with somewhere useful to go next.
SMART OUTPUTS FROM TRAINED
CONVERSATIONAL AI TECHNOLOGY
That is the name.
Useful is the aim.
SOFT CAT is an independent project by Valori. Explore what AI might do next through Horizon, inspect what we have tested in Research, and see what autonomous creation produces in Feral.
Our research follows requests, permissions and actual outcomes. The current studies use scripted inputs. Live-model comparisons are still ahead. Every claim should have evidence you can inspect.
Read our constitution →