Logitration
Logitration — logic-based agent orchestration.
Logitration puts a small logic program in charge of the decisions that must be right, and lets language models do what they are good at: reading what people say and phrasing the answer. Rules and facts decide who does what, when, and with which permissions, so the flow between AI agents and people is predictable instead of improvised, and every answer can come with its reason.
Rules decide, models assist
Deterministic logic makes the decisions. The language model only translates the request into a fixed form and phrases the answer.
Explainable by design
Every conclusion, assignment or refusal carries the rule and the facts that produced it.
Humans in the loop
People and agents share one auditable flow. Nothing is loaded until the person has confirmed it in plain words.
Roadmap
Five stages, from the literature to a fine-tuned small model, each with its honest status today. A plan, not a promise: no dates.
- 1
Literature
Partly done - 2
Logic by levels
Levels 0–3 built and measured on test benches; shell level to be defined - 3
Rule learning
First experiments - 4
Corpus generation
Not started - 5
Fine-tuning
Not started
Where to go next
- How it works — the four steps from a sentence to an answer, and who decides what.
- Evidence — the measurements so far, each with its context and its limits.
- Research — the paper.
- Roadmap — the plan in five stages, with an honest status for each.