ARTENT Genome
The Operating Protocol — a 7-Phase Intent-Driven lifecycle for every AI task.
Traditional
Process-Driven AI
- ×Steps defined upfront
- ×Inputs forced through template
- ×Errors accumulate through pipeline
- ×No learning loop
ARTENT
Intent-Driven AI
- ✓Path generated from task intent
- ✓Routing adapts in real-time
- ✓Errors caught at WF-00
- ✓WF-05 Reflector updates model weights
The 7-Phase Protocol
Interpreter
Receives raw input — text, voice, structured data — and converts it into a normalized Intent Object. No transformation yet; only faithful representation of what was said. Errors caught here cost 1x. Errors caught later cost 100x.
Classifier
Routes the Intent Object to the correct genome, algorithm tier, and LLM roster. This is JITNA in action — matching every task to its optimal processing path without human routing decisions. Sub-100ms target.
Executor
Runs the task across the selected model(s). For HexaCore consensus tasks, this phase fans out to all 7 LLMs simultaneously and collects independent outputs before any comparison or scoring.
Verifier
SignedAI's 8D scoring pipeline runs here. Outputs from Executor are attested, scored across 8 dimensions (Accuracy, Relevance, Coherence, Completeness, Safety, Creativity, Structure, Efficiency), and consensus is established. Outputs below threshold are rejected — not passed with warnings.
Scribe
Formats, signs, and delivers the verified output. Includes audit trail generation (who verified, when, with what models, at what confidence), persistent memory write-back (Vault Genome), and user-facing response formatting.
Reflector
Post-delivery analysis. Every completed task feeds data back into the system: latency, accuracy delta, user acceptance signal, exception types. This is the learning loop that improves G1 (Architect) and updates JITNA's routing weights.
Meta-Phase
The cross-session, cross-genome coordination layer. Aggregates signals from all WF phases across all users, identifies global patterns, and proposes constitutional updates. Updates require human review before they modify genome-level rules (A=0 parameter).
Portable State Container
Every ARTENT session is packaged as a Portable State Container: a complete snapshot of intent history, model choices, verification results, and memory writes that can be transferred across devices in under 2 seconds. This makes ARTENT the first AI agent with true cross-device memory continuity without relying on cloud state management.