You don't need to write any code. Just a table and plain English. The hard parts happen behind the scenes.
What does it do?
It judges your company's requests (expenses, travel…) as ✅ OK / ❌ no, exactly by the rules you set. The AI does not decide on its own. Like a spreadsheet: put values in, get an answer. That's it. Behind the scenes: the same input always gives the same answer (bit-exact), and every decision is recorded (audit).
Press the "Load expense template" button near the bottom of the demo. 80% of the rules are already there. No building from scratch.
A template is a set of common rules (travel, approvals…). You get the 80% ready-made and add only the last 20%.
Try it →In "Spreadsheet mode", change the amount or approver cells. The verdict column instantly shows ✅ valid / ❌ rejected.
Put values in, get the answer. Exactly like a spreadsheet.
In "Write in natural language", write the rule in plain English, then press "Generate S-expr from language".
You never write any code (Lisp). The tool translates.
It automatically checks the converted rule:
If anything is off, the tool auto-rejects it. Wrong rules don't get through. So it's safe.
Press "Golden test" to confirm the template's key rules aren't broken. It tells you if they are.
If OK, press "Publish version". Who set which rule and when is recorded. Made a mistake? Roll back to the previous version instantly.
= you can add freely, but nothing broken goes to production.
Your data and your rules never leave (zero upload). Doubt it? Open the browser's DevTools → Network and use the tool. You'll see nothing is sent.
Summary (this one line is enough)
Type into the table → say the rule in plain English → the tool makes an S-expr and inspects it → finalize.
And you get: the same verdict every time (bit-exact), a record of who decided (audit), all inside your box (zero upload).
A device that seals the atypical judgments born in conversation into deterministic executables, on the spot ― driven by nothing but a table and plain English.
Open the demo and try it →* For complex sentences the light generator can't handle, in production an AI placed inside your company (local Gemma4) does the translation. Either way, after conversion the same inspection and audit apply.
* Real production (canonical records, org-wide versioning) requires membership. This is the free demo.