Deterministic and Agentic AI in One Hybrid Core

Features

1. 

   04.1

   Typed Handoffs

   Every step writes into a declared type, model steps included. The wiring is checked against those types on save, and every mismatch is named.

   • Declare what each step returns.
   • Make model steps return typed data.
   • Catch wiring mistakes on save.
   • See every mismatch by name.
2. 

   04.2

   Confidence the Flow Acts On

   A decision step returns a probability, a choice or a score. The flow accepts above one mark, rejects below another, and sends the middle to a second check.

   • Set your own confidence thresholds.
   • Auto-accept confident answers.
   • Reject answers below your threshold.
   • Send borderline cases to second check.
3. 

   04.3

   Rule-Based Routing

   A branch step routes each input by rule, such as file type, host or a field. A guard can refuse a call with the flow’s own message.

   • Route inputs by file type.
   • Route by host or field value.
   • Send each case down its path.
   • Refuse bad requests with your message.
4. 

   04.4

   Bounded Loops

   Refine-until loops stop on a condition the flow states. An iteration cap either fails the run or ends it with a warning.

   • Refine output until a condition holds.
   • Cap how many times loops run.
   • Fail a run at the cap.
   • End capped loops with a warning.
5. 

   04.5

   Per-Item Judgement, Reusable Parts

   A list fans out so each item gets its own decision, then merges back. A proven flow is called as one step, and the validator catches cycles.

   • Judge each list item separately.
   • Merge results back into one.
   • Call proven flows as single steps.
   • Build larger processes from proven parts.
   • Catch flows that call themselves.