Softanza
StartFrançais
Softanza
How a machine reads Softanza

For agents

This section also addresses an AI agent reading this site. Softanza is designed to be read by a machine as much as by a human.

Ask the library

The library documents itself. An object answers Ask() with its own methods, explains each with ExplainMethod(), and says how with HowTo(). An agent does not need to guess a name.

Softanza
? Q([ 1, 2, 2 ]).Ask("how do I remove duplicates")
Output
Unique
1.10
remove duplicates / unique (engine-backed)
UniqueCS
...
RemoveDuplicates

run on 2026-10-01 at 01:40

Every call run, and measured against the library's own recipes

Declare yourself in an agent file

An agent is a file, judged at load: what it covers, the reversibility class of its acts, the execution posture of every function it calls. The court refuses in fixed sentences, the same at both doors.

A bank analyst declares a stock-watcher agent,
first without saying what it covers:
-> refused / [pia-coverage @ coverage]
   an agent must say WHAT IT COVERS.
...then with its coverage stated:
-> The court judged your declaration,
   and every promise of the exercise was kept.

run on 2026-09-30 at 23:10, demo, scene 7

Speak the grammar, not general code

A schema already compiles to a constraint grammar: a local AI model whose sampler is held to it can only emit valid answers. That is the platform's contract C9: structure kills malformedness, never falsehood. Emitting such a grammar for every declared language is the next step, decided as the command stz grammar.

SOFTANZA_INTELLIGENCE_ARCHITECTURE.md(opens in a new tab)