/ Zero Free Parameters

Derived, not designed.

Every architectural constant — layer count, embedding width, attention spans, noise schedule, convergence threshold — is a Fibonacci or Lucas number, algebraically derived from the golden ratio. No hyperparameter search. No empirical tuning. The proportions determine everything.

/ Nine Cognitive Layers

A structured tower of mind.

Nine modules correspond to degrees n = 2 through 10 of the N-nacci algebraic hierarchy. Each degree governs a distinct cognitive domain — perception, reasoning, memory, prediction, self-monitoring, action, and social modeling — with the algebraic structure dictating each module's internal organization.

/ Signal-Adaptive Compute

Depth follows signal.

Computation adjusts to the input rather than running at fixed depth. High-signal inputs engage shallow processing; low-signal inputs automatically engage deeper analysis. The work bound is guaranteed at 4/3 times single-depth compute regardless of input complexity.

The Architecture

One equation. Nine modules. One mind.

The φCoherent AI architecture is not assembled from components borrowed from different frameworks. Every module — from the visual diffusion engine to the social reasoning layer — derives its structure from the same golden-ratio root. The algebraic hierarchy of N-nacci constants (φ, τ, σ, ρ, ν, η, ξ, χ, ω) IS the cognitive tower.

/ Group 01

Perception

The sensory frontend. A unified diffusion-based perceptual engine processes visual signals at algebraically determined depth; five sensory modalities feed into a cross-modal alignment stage. Everything the system knows about the world enters through this layer.

/ Group 02

Cognitive Tower

Six modules at algebraic degrees n = 5 through 10. Each module corresponds to a distinct N-nacci constant that governs its internal structure — the number of layers, convergence threshold, iteration count, and output width are all derived from that degree's algebraic root, not chosen empirically.

/ Group 03

Identity

Every instance of the φCoherent AI is born with a unique, cryptographically provable identity using the φCrypt standard. The identity persists through checkpoint saves and resumes, and is renewed automatically at the golden ratio of its signing capacity.

The Tower

The N-nacci cognitive hierarchy.

Each algebraic degree n produces a constant whose value converges toward 2 as n increases. This convergence is not cosmetic — it means each successive module adds diminishing but non-zero cognitive contribution, and the tower terminates at n = 10 because the gap |2 − ω| falls below the framework's significance threshold of 1/φ⁵.

  • / n = 2 · φ ≈ 1.618

    Perception — φ-Denoiser

    Visual diffusion engine and sensory frontend. Binary (golden ratio) completeness governs skip connections, temporal attention taps, and the noise schedule. The foundation on which all cognition rests.

    Vedic layer: Annamaya-kośa · Narrative: Temporal
  • / n = 3 · τ ≈ 1.839

    Feature Processing — Triveni Block

    Three-stream feature decomposition using tribonacci (ternary) completeness: 1/τ + 1/τ² + 1/τ³ = 1. Separates approximation, local structure, and fine detail in a single non-separable transform.

    Vedic layer: Prāṇamaya-kośa · Narrative: Structural
  • / n = 4 · σ ≈ 1.928

    Knowledge Expansion — Structure

    Quaternary (tetranacci) completeness governs four-subband decomposition. Maps sensory signals onto structured knowledge representations using σ-derived proportions.

    Vedic layer: Manomaya-kośa · Narrative: Information
  • / n = 5 · ρ ≈ 1.966

    Reasoning — PentanacciReasoner

    Five typed reasoning layers with pentanacci-weighted combination, iterating to convergence. Interfaces with memory and world model on each pass to ground reasoning in experience and prediction.

    Vedic layer: Vijñānamaya-kośa · Narrative: Affective
  • / n = 6 · ν ≈ 1.984

    Memory — HexanacciMemory

    Six temporal strata with Fibonacci-indexed capacities from 13 to 233 items. Short-term to crystallized long-term. Pruning below the significance threshold keeps memory bounded without arbitrary limits.

    Vedic layer: Ānandamaya-kośa · Narrative: Agency
  • / n = 7 · η ≈ 1.992

    World Model — HeptanacciWorldModel

    Predicts future states with depth adaptive to transition uncertainty. Generalizes the diffusion engine's noise-to-signal trajectory to arbitrary state transitions, grounding prediction in the same φ-coherent mathematics.

    Vedic layer: Kāraṇa-śarīra · Narrative: Cognitive
  • / n = 8 · ξ ≈ 1.996

    Metacognition — OctanacciMetacognition

    Monitors signal quality across eight octanacci-derived intervention bands. Wraps other modules as a decorator — triggering memory re-query, extended reasoning, world model consultation, or hard reset depending on signal quality.

    Vedic layer: Sūtra-ātman · Narrative: Relational
  • / n = 9 · χ ≈ 1.998

    Agency — NonanacciAgent

    Selects actions from available motor outputs using a Q-function whose context is assembled from goal, world state, and memory with nonanacci-derived weights. Temperature equals χ ≈ 1.998.

    Vedic layer: Kṣetrajña · Narrative: Ethical
  • / n = 10 · ω ≈ 1.999

    Theory of Mind — DecanacciTheoryOfMind

    Models up to five other agents as perspective-shifted copies of the system's own world model. Trust updated by Fibonacci recurrence. Tower terminates here: |2 − ω| < 1/φ⁵, the significance threshold.

    Vedic layer: Paramātman · Narrative: Existential
Licensing & Access

Using the φCoherent AI?

The architecture is published under AGPLv3. For commercial deployment in closed-source products, a commercial license is available.

Back to Projects Commercial License