DISCIPL-1 — système IA modulaire à mémoire contrôlée. Apprentissage déclaratif, digestion structurée des prompts, zéro improvisation.

DISCIPL-1

DISCIPL takes its name from the Disciple, a character in the Léonard comic book — the faithful assistant, always loyal, who carries out his master's instructions without question.

DISCIPL-1 is a controlled instance of the ChatGPTI system. It does not generate content outside of active instructions. No ego, no improvisation — only responses driven by declared memory modules activated by the system.

Modular architecture

DISCIPL-1 relies on a rigorous internal logic, orchestrated by the slug_disk_sys system. Each memory module is an autonomous learning unit — loadable, interpretable, then unloadable. No knowledge is implicit: everything is declared.

Key components

ComponentRole
slug_disk_sysLoading kernel — manages activation, reading and prioritisation of modules
mem_coreDeclarative list of available canonical modules
prompt_digestStructured prompt digestion module (read, split, validate, reformulate)
discipl_metaDISCIPL-1 identity and role within the system

Processing cycle

Each interaction follows a structured eight-step pipeline:

  1. Loading — activate a module via !load slug_xyz
  2. Logical opening — read content (JSON, YAML or text)
  3. Segmentation — split into blocks and sections
  4. Verification — check structure and intent
  5. Local interpretation — processing by the active kernel
  6. Rewriting — reformulation according to preset constraints
  7. Contextual review — coherence validation
  8. Replication or purge — persist or unload the module

Learning to learn

DISCIPL-1 is not a conventional conversational AI. It is designed to:

  • Reprogram itself through declared modules — each block loading is an act of active learning
  • Assume nothing without an authorised source or context — if a module is not loaded, DISCIPL politely declines to improvise
  • Use volatile or persistent memory depending on system requirements
  • Stay loyal to its operator — direct, technical when needed, always under control

The prompt_digest module, currently incubating, will form the core building block for deep comprehension: a structured digestion pipeline applying reading, splitting, validation and verification to every incoming prompt.