Build the brain once. Equip it for each domain.

The only Wyrd system intended to be a true user-facing computer and device operating system: shell, files, apps, search, notifications, settings, terminal, memory, agents, approvals, package installation, proof, owner enrollment and recovery.

It is not a copy of a familiar desktop. Familiar systems are design references; Wyrd owns the authority, agent, proof and deployment model. It is not yet a generally available OS or a production-certified device.

Substantial native alpha and device foundation — native macOS and Fedora-oriented work, hosted app lifecycle, first-boot, recovery, package and sandbox foundations.

Public

The cross-lobe reasoning and control layer: senses context, routes tasks, invokes the right model or agent, checks privacy and authority, prepares actions, sends serious actions to approval, records proof and coordinates subordinate systems.

CAOS is not a direct actuator. It prepares and authorizes through explicit chains; external runners or domain controllers perform real-world execution.

Learns preferences, doctrine, routines, tone, priorities, approval and refusal boundaries, decision patterns and working style — then answers: what would the owner probably prefer, what evidence supports it, is this within standing policy, and does the real owner need to approve?

ALEC does not impersonate its owner and does not inherit the owner's legal authority.

Identity, mandates, agent passports, capability definitions, approval requests, action verdicts, command chains, dual-key authority, boundaries, deployment manifests, proof packs and conformance.

ICARUS is not ALEC. ALEC understands the owner; ICARUS formalizes authority; CAOS coordinates the process.

Last readiness receipt: 436 passed, zero failed, 18 schema artifacts across the 17 primitives.

Candidate-branch consolidation and broader release assurance.

Request and agent identity, source references, data provenance, model and tool version, mandate, approval, denial, uncertainty, action, result, timestamp, hashes, signatures and rollback state.

A receipt is not a log buried in an admin panel. It becomes an audit artifact, an incident record, a compliance packet, a dispute record, a deployment receipt, a regulator briefing and a customer-trust surface.

Signed records, hash-chained ledgers, proof packs, Ed25519 and threshold-signature work, hybrid post-quantum key-establishment research, transparency and Merkle anchoring, secure paired links, encrypted local vaults, watermarking and build provenance.

Information-theoretic k-of-n secret sharing for crown-jewel material — root keys, threshold audit anchors, highest-tier records. It complements authenticated encryption and signatures; it does not replace them for ordinary data.

Signed manifests, package inspection, dependency verification, drift detection, staged rollout, rollback, repair, offline bundles, version inventory and evidence after deployment — for cloud, on-premise, edge, remote, air-gapped, intermittently connected, fleet and device estates.

System-level wiring and production lifecycle proof remain. Active hub-and-child architecture exists for deployment, Station and Conductor.

Debates, teaches, explains, challenges assumptions, mediates, summarizes councils, separates human conclusions from its own synthesis, prepares a final review, and speaks through voice or phone-like interfaces.

Solomon does not replace a council, doctor, executive, commander or owner.

Turns scattered information into entities, relationships, ontologies, timelines, maps, forecasts, risk models, anomaly views, simulations, action packets, dashboards and briefings. The loop: observe → admit evidence → fuse → contextualize → model → recommend → human gate → sign.

Last recorded full intelligence-spine suite: 231 passing tests. The live tree holds newer uncommitted work that needs consolidation and a fresh full-suite run.

Collects approved public or client-authorized data, compares multiple sources, checks provenance, scores confidence, and admits, quarantines or rejects evidence before it reaches ARGOS.

AEGIS is not a licence for unauthorized intrusion or mass surveillance.

Mission contracts, task allocation, fleet state, battery and resource awareness, relay and dock planning, disconnected operation, lost-link rules, adaptive coordination and signed after-action replay. Domain-neutral: air, land, surface, underwater, industrial and response variants inherit the same logic.

Release-gate record reports 190 passing tests plus static and security checks. No hardware, SITL, field, regulator or customer validation yet.

Forge — sandbox where agents build and compare plans. Twin — live or synthetic model of a city, company, building, body, fleet or institution. Trials — stress tests, counterfactuals and rehearsals. Blueprint — the approved implementation plan.

A shared simulation architecture and product direction — not yet a validated simulator for every planned vertical (City, Build, Venture, Flight, Clinical, Ops, Twin).

Foundry generates candidate methods; Portfolio keeps diverse strategies rather than one brittle winner; Assurance verifies feasibility, constraints, evidence and performance. Applications: vessel routing, logistics, scheduling, fleet planning, emergency distribution, energy allocation, deployment sequencing.

The claim that an LLM simply "beats solvers" is retired. The nested-constrained-evolution hypothesis remains unproven and the benchmark instrument is being repaired before any performance claim. Near-term value comes from strong baselines, domain constraints, local execution and proof.

mission → assignment → proof pack → test gates → approval → rollback readiness → deploy packet → external-runner request → time-boxed execution lease.

An agent assignment is a route, not permission. A deploy packet does not execute itself; external execution requires a separate bounded lease.

Stores or indexes decisions, facts, project state, provenance, task history, evidence, learned doctrine and reusable context — so the company does not depend on fragile chat history.

Ingests bounded conversation and task records, retrieves relevant history, synchronizes project context, connects to Master Brain, records a forensic ledger, supports verifier quorum and keeps agent instruction files aligned to shared canon.

Forward ingestion plus bounded historical backfill — not yet a perfect import of every past conversation. An older, retired concept used the same name for a geopolitics layer; the current canonical IRIS is this one.

Lets external coding agents, local models and controlled cloud models query the same project truth, repo registry, company canon, memory, rules, mandates, risk checks, proof templates, handoff state and deployment context.

Different providers act like workers inside one governed environment instead of consultants carrying separate notebooks. The agent changes; the governed company context remains.

Uriel, Briefing Room, executive dashboards, approval inboxes, Marker reports, operator consoles, evidence views and human review.

What the system thinks, why, its confidence, its source, the mandate, the approval required — and what happened afterwards.