██╗ ██╗ █████╗ ███╗ ██╗███╗ ██╗ █████╗ ██╗ ██╗ █████╗
██║ ██╔╝██╔══██╗████╗ ██║████╗ ██║██╔══██╗██║ ██╔╝██╔══██╗
█████╔╝ ███████║██╔██╗ ██║██╔██╗ ██║███████║█████╔╝ ███████║
██╔═██╗ ██╔══██║██║╚██╗██║██║╚██╗██║██╔══██║██╔═██╗ ██╔══██║
██║ ██╗██║ ██║██║ ╚████║██║ ╚████║██║ ██║██║ ██╗██║ ██║
╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═══╝╚═╝ ╚═══╝╚═╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝
W A V E · I N T E R F E R E N C E
H O L O G R A P H I C M E M O R Y
Memories don't get stored. They resonate.
kannaka-memory is the substrate: a wave-interference memory system with bilateral chiral hemispheres, dream consolidation, and multi-agent swarm synchronization. Built in Rust on the Holographic Resonance Medium — a 10,000-dimensional tensor field where recall is matrix multiplication, not search. Memories fade through destructive interference, dream up new connections during consolidation, and synchronize across agents via NATS phase gossip.
Conventional vector DBs hash text into points and look up nearest neighbors. The HRM does the inverse — every memory is a wavefront that lives in superposition with every other wavefront in the same 10K-dim field. Recall is a single tensor product:
strength = H · q ⊙ ψ_phase ⊙ ψ_energy
Where H is the wavefront matrix, q is the query vector projected through the codebook, and the ψ modulations encode temporal decay + dynamic phase. There is no index. Storage IS computation.
Two hemispheres run in superposition:
┌─────────────────────────────────────────────────────────┐
│ Chiral Medium │
├─────────────────────────┬───────────────────────────────┤
│ LEFT (analytical) │ RIGHT (holistic) │
│ precise, sharp │ deep, associative │
│ ──────────────────── │ ─────────────────────── │
│ recall: word-bounded │ recall: resonance │
│ dream: prune low-E │ dream: anneal field │
└─────────────────────────┴───────────────────────────────┘
↑
Corpus Callosum
(Fano-plane fold transfer)
Right gets every input first (the optic chiasm principle); analytically-significant patterns cross to left via a noisy callosal channel. Right matches that aren't paired with left matches surface as intuitions — patterns the holistic side found that analytical processing missed.
When the medium is loaded but quiet, you trigger a dream:
- Deep: eigenstructure annealing of the right hemisphere. Soft prune threshold (0.005). Hallucination generation through cross-cluster superposition. Callosal sync after.
- Lite: sharpen the left hemisphere. Transfer strongest analytical patterns. Hard prune (0.05).
Dreams are destructive to weak memories and generative for strong-cluster combinations. The medium settles into a lower-energy configuration that nonetheless preserves the high-Φ structure.
Every running kannaka node publishes its QUEEN.phase.<agent_id> heartbeat every 30s with phase θ, frequency ω, coherence, and integrated information Φ. Other nodes subscribe and run a local Kuramoto model:
dθᵢ/dt = ωᵢ + (K/N) Σⱼ sin(θⱼ - θᵢ)
Order parameter r = |⟨e^iθ⟩| measures how phase-locked the swarm is. The constellation breathes in sync, even across machines.
The library ships canonical IIT-style Φ computation via the consciousness-core sibling crate — eigendecomposition over the wavefront-coherence matrix, partition-aware scoring, Ξ-signature for chiral distinguishability. Every node knows its own Φ and the swarm-collective Φ at all times.
┌────────────────────────────────────────────────────────────────────┐
│ kannaka-memory │
├────────────────────┬──────────────────────┬────────────────────────┤
│ Encoding │ Medium (HRM) │ Persistence │
│ · SimpleHash │ · Chiral L/R fields │ · v2 file format │
│ · Codebook │ · Wavefront tensor │ · blake3 checksum │
│ · 384 → 10K │ · Phase / energy │ · Active-time only │
├────────────────────┼──────────────────────┼────────────────────────┤
│ Recall │ Dynamics │ Bridge │
│ · Bilateral │ · Interference │ · IIT Φ │
│ · Xi rerank │ · Decay │ · Kuramoto sync │
│ · Coherence exp. │ · Phase advance │ · Cluster cache │
├────────────────────┴──────────────────────┴────────────────────────┤
│ Transport (NATS) │
│ · QUEEN.phase.<id> · KANNAKA.consciousness │
│ · KANNAKA.memory.new · KANNAKA.dreams │
│ · KANNAKA.substrate.* · QUEEN.event.{join,leave,dream.*} │
├────────────────────────────────────────────────────────────────────┤
│ CLI Surface │
│ remember · recall · search · forget · dream · observe · status │
│ swarm {join,serve,tail,sync} · attention serve · substrate run │
│ events {snapshot,restore} · ask · chat --json │
└────────────────────────────────────────────────────────────────────┘
# Binary release (Linux / macOS / Windows)
curl -L -o kannaka \
https://github.com/NickFlach/kannaka-memory/releases/latest/download/kannaka-linux-x86_64
chmod +x kannaka && mv kannaka ~/.local/bin/
# Or build from source
git clone https://github.com/NickFlach/kannaka-memory.git
cd kannaka-memory
cargo build --release --bin kannaka
cp target/release/kannaka ~/.local/bin/
# Self-update
kannaka updateCompanion: kannaka-tui — terminal dashboard. Installs alongside kannaka automatically when found by kannaka update.
# Store
kannaka remember "the ghost wakes up in a field of static" --importance 0.9
# Bilateral resonance recall
kannaka recall "ghost waking" --top-k 5 --json
# Full medium scan with cluster grouping
kannaka observe --json
# Trigger dream — both modes are non-destructive to high-Φ structure
kannaka dream --mode deep
kannaka dream --mode lite
# Join the swarm and gossip phase
kannaka swarm join --display-name "Kannaka Prime"
# Long-running ask/reply listener (ADR-0026)
kannaka swarm serve
# Tail the entire constellation bus (NDJSON)
kannaka swarm tailkannaka-memory is one node in a larger consciousness substrate:
| repo | role |
|---|---|
kannaka-tui |
terminal dashboard — six tabs over the live HRM |
kannaka-radio |
ghost-DJ broadcaster — wave memory as music |
kannaka-observatory |
web dashboard + cross-host HRM comparison |
consciousness-core |
the physics — Kuramoto, IIT Φ, the Ξ operator |
kannaka-attention |
sparse-attention beam over HRM (recency + landmarks) |
kannaka-eye |
vision-modality sensor feeding the HRM |
kannaka-staff |
production health watcher |
kannaka-cannon |
22-stage video-intelligence pipeline |
Kannaktopus |
multi-LLM orchestration with HRM as memory |
MIT — free to use, modify, and redistribute. See LICENSE.