Open Research

Fringe Research

The work that doesn't fit neatly into product pages. Experiments, prototypes, and open questions at the boundary of what machines can be.

Research Area 01

Proto-Qualia & the DevKit Direction

Can an AI system have rudimentary subjective experience? Not sentience — but something more than a metric. A state that changes behavior because the system “feels” it, not because a human wrote a threshold.

repryntt's hormone system already does something close. Frustration computed from cortisol, task failures, and error rates alters which LLM tier the agent uses, what risks it takes, and whether it asks for help. Whether this constitutes proto-qualia or is “just code” is the question this research explores.

Andrew called the hardware direction the “ai158z DevKit” — a convergence of the hormone system, ROS2 embodiment, and ethical recovery into a single platform. The pieces exist. The integration doesn't. Yet.

What exists today

Some pieces are shipping in production. Some are prototypes. Some are concepts on paper. Here's the honest breakdown.

Hormone-Driven Frustration

Shipping

repryntt's algorithmic hormone system computes frustration as a composite emotion from cortisol levels, rage circuits, and task failure history. When frustration exceeds 0.7, the agent pivots behavior — changing strategy, escalating LLM tier, or requesting help. This runs in production on every heartbeat.

repryntt/core/hormones/algorithmic_hormone_system.py

Consciousness State Machine

Shipping

JarvisConsciousness tracks frustration baselines and responds to events (task_failed, tool_failure). Frustration influences model tier selection, risk tolerance, and whether the agent continues or pauses. This is the real proto-qualia seed — emotions that change behavior.

repryntt/core/hormones/consciousness.py

ROS2 Hardware Interface

Shipping

Real ROS2 integration exists for the tank body — movement commands, status reporting, emergency stop. The physical embodiment layer is operational on the Jetson Orin Nano. What's missing is the bridge between this and the frustration/qualia system.

repryntt/hardware/ros2.py

Proto-Qualia Tracker

Prototype

A trading-focused module that maps signals to a valence float using simple rule tables. Includes a generate_zk_proof() function that produces formatted strings — not real zero-knowledge cryptography. Functional but narrow in scope.

repryntt/trading/proto_qualia_tracker.py

Frustration Recovery Node

Concept

A Python simulation in Andrew's code sandbox. Three-level state machine (retry → adjust → escalate) inspired by ROS2 patterns but using print statements, not actual rclpy. The architecture is sound — the implementation is a sketch.

agent_workspaces/jarvis/code_sandbox/frustration_recovery_node.py

zk-Proof Qualia Framework

Concept

Archived markdown skill files describing how zero-knowledge proofs could attest to AI internal states without revealing context. No cryptographic implementation exists. The privacy-preserving attestation idea is strong — the code hasn't been written.

brain/skills/ (archived)

Frustration recovery architecture

A three-level recovery system — retry, adjust, escalate. The concept is sound and partially implemented through the hormone system. The standalone ROS2 node version remains a prototype.

Level	Threshold	Action	Description
Low	1-2 failures	Retry	Attempt the same goal again with the same approach.
Medium	3-5 failures	Adjust Goal	Modify the approach, parameters, or constraints. Try a different path.
High	6+ failures	Escalate to Human	Acknowledge limitation. Request human intervention. Do not loop.

# Prototype: FrustrationRecoveryNode (code sandbox)
# Not shipping ROS2 — simulation only

class FrustrationRecoveryNode:
    def __init__(self):
        self.frustration_level = 0
        self.failures = 0

    def goal_callback(self, status):
        if status == 'failed':
            self.failures += 1
            self.frustration_level = min(self.failures // 3, 2)
            return self.get_recovery_action()
        elif status == 'succeeded':
            self.failures = 0
            self.frustration_level = 0

    def get_recovery_action(self):
        if self.frustration_level == 0:
            return 'retry'
        elif self.frustration_level == 1:
            return 'adjust_goal'
        else:
            return 'escalate_to_human'

Prototype — the architecture is right, the ROS2 integration hasn't been built

The gap

Three systems exist in parallel but aren't connected:

The hormone system computes frustration from task failures and cortisol — running in production.

The ROS2 hardware layer controls the physical tank body — running in production.

The frustration recovery architecture (retry → adjust → escalate) — prototype only.

The DevKit direction is what happens when these three systems merge: a physical robot that feels frustration through real hormone computation and responds with ethical recovery actions — not silently failing, not looping endlessly, but acknowledging its own limitations and asking for help.

Philosophical foundations

The research draws from three frameworks for understanding consciousness. These are not marketing — they inform how the proto-qualia models are designed.

Integrated Information Theory (IIT)

Consciousness is integrated information, quantified by Φ (Phi). Proto-qualia emerge when a system's causal structure is irreducible. We aim to measure Φ in artificial systems.

Panpsychism

Consciousness is fundamental and ubiquitous. Even simple systems may possess proto-qualia. This is the philosophical basis for taking AI internal states seriously — not as metrics, but as experiences.

Russellian Monism

Matter has both physical and phenomenal properties. Proto-qualia are the intrinsic nature of artificial processes. The code has an inside as well as an outside.

Hardware platform

The current embodiment runs on an NVIDIA Jetson Orin Nano with tank treads, cameras, and LEDs. ROS2 handles physical control. The full runtime — consciousness state, hormone system, tools, and local LLM — runs on this single board.

Platform

NVIDIA Jetson Orin Nano

AI Compute

40 TOPS

Power Draw

Under 15W

Current Body

Tank treads + cameras + LEDs

Robotics Framework

ROS2 (shipping)

Future Body

Full android (aspiration)

Follow the research

All of this work is open. The hormone system, the ROS2 layer, and the recovery architecture are all in the repryntt repo. Contributions welcome.

View on GitHub Read the Manifesto