Omega is the decision orchestrator at the heart of every HYVE agent. It runs a deterministic decide() function — same input, same output, every time, under 200 nanoseconds per rule — over a topologically-sorted multi-organ topology, atop the two-tier key hierarchy and GF(256) recovery that make the whole architecture sovereign. Every verdict is replayable in Observatory.

Connect every LLM you already pay for AND train your own from scratch — Anthropic, OpenAI, Azure, 30+ frontier hosts, local engines, quantum compute providers, and biological-substrate backends — through one OS. Fine-tune existing models on your private data, or train new models end-to-end on your hardware. You bring the keys, HYVE never sees your password.

Five algorithmic patterns route across ten quantum compute providers — the first agent OS that treats quantum hardware as a first-class compute target alongside classical and biological substrates.

Three-dimensional graph of organ nodes on a Fibonacci-sphere distribution, lightning polylines for cell deliveries, color-coded by topic class across at least 8 categories. Every decision, memory write, and bus event is replayable.

Every byte on the substrate flows through a 512-byte uniform encrypted cell — ChaCha20-Poly1305 AEAD, ML-KEM-768 key exchange, Ed25519 signatures, blind routing tokens. Built for the day quantum computers break the rest of the internet.

Camera and microphone cut at the kernel module level, below userspace, below the browser. No software can override the hardware kill.

Eleven static analysis detectors across five languages: secrets, weak_crypto, sql_injection, command_injection, deserialization, insecure_random, unsafe_audit, panic_audit, xss, regex_dos, ssrf.

Augur audits its own source on every commit — the security tool that catches itself before catching anything else.

A tarpit organ that drips fake banners to attackers for hours, a Tor controller organ with pluggable transports (obfs4, meek_lite, snowflake), and a sentinel kill organ for cross-platform process termination — all audit-logged.

PIN, password, gesture, face biometric, fingerprint biometric — at least one combination required to unlock. A second 'panic' PIN at the unlock prompt marks the data partition for irreversible secure wipe on next reboot.

Sovereign on-device identity: X25519 master keypair, Shamir 3-of-5 social recovery, HKDF-derived scoped tokens. No cloud account required, ever.

Per-chunk Brotli + AES-256-GCM encryption with an Ed25519-signed root hash you can verify offline, plus SQLite-indexed metadata; one vault unlock decrypts every archive. (NOT yet in the shipped build: Reed-Solomon GF(256) parity shards + Particle-Swarm-Optimized layout for partial-corruption survival — the designed next layer.)

Messages exceeding 387 bytes are split via Shamir K-of-N secret sharing applied per byte position over GF(256). Any K cells reconstruct the original; any K-1 cells reveal nothing.

Multi-factor unlock with length-capped inputs, parallel Argon2id derivations without short-circuit, bitwise-AND combination, and uniform error message on failure — defends against side-channel timing attacks.

Per-conversation epochs with chain-key rotation via HKDF-SHA256 and zeroized prior keys. Compromising current state never compromises past traffic.

Peer-to-peer LAN file transfer organ — mDNS discovery, TLS-fingerprint pinning, no relay server, up to 8 GiB per file. Your files never leave your network.

A standalone binary that exposes Omega, Narrator, Augur, Trust, and Leaper as native MCP tools — call HYVE organs directly from Claude Code, Cursor, Zed, and Continue.dev.

Per-IDE skill manifests (.claude/skills, .cursor/rules, ~/.continue/config.json) so operators invoke HYVE organs from inside their AI coding harness without leaving the editor.

Six plugin types with manifest-based install on ports 7900–7999 — Studios, organs, themes, recipes, surfaces, agents — all extendable.