Intelligence you can govern.
Wyrd Group builds governed, local-first and cloud-optional AI systems that let people, institutions and machines use intelligence without surrendering authority, privacy, resilience or proof.
Local when sovereignty matters. Cloud when useful. Always governed.
Intelligence without authority.
Models keep getting better at producing answers. Almost nothing decides who asked, what the agent may do, which data it may touch, when a human must sign, what can be revoked, and what can be proved afterwards. That missing layer is what stops capable AI from entering the institutions that carry the most consequence.
Capability, unattributed.
- An answer with no admitted source
- An action with no mandate behind it
- A request that leaves the perimeter unredacted
- No approval gate before a consequential step
- A log, not an evidence packet
- Nothing to deploy, repair, roll back or revoke
- Nothing left working when the network disappears
The institutional operating layer.
- Institution-owned identity and local memory
- Role and mandate boundaries, signed
- Redaction before anything leaves the perimeter
- Approval gates and dual-key authority
- Proof receipts as a product surface, not a log
- Deployment, repair, rollback and revocation
- Local fallback and continuity when links drop
Not "cloud AI is bad." Cloud providers supply intelligence; Wyrd supplies the institutional operating layer that makes that intelligence governable. That lets Wyrd complement frontier and open-source providers rather than depend on any one of them.
One governed path, from evidence to receipt.
Wyrd is a reusable software platform, not a chatbot and not a loose collection of products. Every serious action travels the same path — whatever the domain.
Wyrd Labs creates the software. The divisions turn it into market products.
One software-production division and three commercial divisions. Pick a unit to see which shared blocks it inherits and what those blocks are assembled into.
What is working, what is being integrated, what still has to be proved.
The company contains real software, large integrated alphas, pilot candidates, research prototypes, hardware programs and long-term options. Every entry on this site carries the rung it actually sits on.
Running code with test or runtime evidence
- ICARUS primitives
- Proof Spine components
- Perfect-Secrecy Vault
- Themis
- Wyrd MCP · Master Brain
- Conformity Lab
Components work together; release work remains
- T-AOS device foundations
- ALEC Personal Cortex
- CAOS governance cycle
- Solomon client code
- ARGOS intelligence spine
- PEAOS engineering lobe
Coherent enough for a gated design-partner pilot
- Nestar Agentic OS
- Raphael clinician pilot
- TrueSight authentication
- RADOME continuity demo
- Anchorline site pilot
- CTOS read-only workflow
Hypotheses under test, or integration not yet done
- NINA method foundry
- Substrate Hypothesis
- DAEDALUS simulation
- Leviathan integration
- Phaeton robotics
- NOMAD design study
Individual receipts, not one inflated total.
A point-in-time engineering snapshot per system, with what the evidence does not establish stated next to it. We do not add the test counts together.
Innocence Translation.
Not "turn dangerous technology into a civilian product." It is the discovery method: protect the question, then let evidence decide whether it survives.
Innocence asks the question. Expertise tries to kill it. Proof decides.
Pilots start with shadow mode, not your live systems.
Wyrd does not need access to live critical systems on day one. Pilots begin with historical, synthetic or non-critical data; observation comes before recommendation, and recommendation before any bounded control.
Every quarter, Wyrd judges AI — then judges itself.
Three standing functions exist to make power face judgment before leadership decides, and to make that judgment public on a schedule.
Wyrd's governance and proof architecture is designed for high-consequence environments — including restricted programs that are named and placed on the restricted & partner register. The operational content itself is excluded from all public and general partner materials, in every tier, and disclosed only through authorized, controlled channels — lawful customer authority, licensing, export review and independent legal and safety control:
- Targeting, weapons or kill-chain logic of any kind
- Intrusion, persistence or destructive cyber methods
- Unauthorized surveillance or ways around civil restraint
- Covert control of a client's machine
- Unlicensed payment, banking, exchange or currency claims
- Medical diagnosis or treatment claims without the required validation
The programs exist and are named. Their operational content stays out of public and general materials entirely, released only through those authorized channels — never against curiosity.
Wyrd builds intelligence that can act without escaping ownership, authority or consequence.
Fund the proof, not the promise. We are selecting design partners for controlled pilots.