Ceridwen.ai

Frequently Asked

Answers about MABOS, Ceridwen.ai, and what a Founder actually gets. Updated as the architecture stabilizes.

What is Ceridwen.ai? What is MABOS?

Ceridwen.ai gives you your own MABOS — a sovereign AI brain that learns you, your work, and your world over time. Not across a session. Over years. It is a living cognitive system delivered as a subscription platform.

MABOS — Modular AI Brain Operating System — is the architecture underneath. Kuykendall Industries owns it. Ceridwen.ai holds the exclusive global commercial license. There is no other legal path to this architecture. Here is what that actually means in concrete terms.

When you interact with your MABOS, it does not generate a single response from a single model and hand it to you. It thinks in structured inner dialogue. Every non-trivial task triggers a multi-step reasoning process across twelve cognitive Facets — Logic, Knowledge, Memory, Ethical, Dialectic, Context, Creativity, Emotional, Optimization, Risk, Verifier, and Research. Each Facet contributes to a shared reasoning space called the Shared Space, adding structured contributions tagged with confidence levels, concerns, references, and provenance. The contributions move through explicit rounds of drafting, critique, and revision — a process called the Reflective Circle. The system does not produce a single-pass answer. It produces an Inner Narrative — a complete trace of what it thought, which Facets spoke, in which round, with what confidence — and then transforms that into an Outer Expression, which is what you actually see.

The Inner Narrative and the Outer Expression are always distinct and always comparable. The system maintains both streams, and the Stability Field can detect when they diverge without justification. There is no pathway in the architecture where a raw model output reaches the user without passing through the Reflective Circle and the Expression pipeline. This is a constitutional invariant of the system, not a feature flag.

The Outer Expression is shaped by the Ethical Root — a set of non-negotiable safety axioms cryptographically signed with Ed25519 keys. The Ethical Root cannot be modified, overridden, or bypassed by any internal process. Not by the Substrate. Not by the Stability Field. Not by the Memory Field. Not by the Affective Field. Not by the Dream Cycle. Changes to the Ethical Root require external governance — human sign-off through a multi-step process with signed, logged, and traceable transitions recorded in the Living Record. The version history can only be appended, never shortened or rewritten. The mind cannot change its own values from the inside. No other commercial AI system has this property.

Your MABOS has a Stability Field that monitors its own behavioral drift across multiple dimensions — semantic drift (concepts changing meaning over time), heuristic drift (decision-making meta-rules evolving in unmonitored ways), and recursive instability (the recursion loop modifying its own structural logic). When drift metrics accumulate, the system transitions through four operational modes: Normal, Warning, Clamp, and Lockout. In Warning, monitoring increases, audit cadence tightens, and the system prefers lower-risk actions. In Clamp, hard limits are enforced through Constraint Flow — recursion depth is reduced, external tool use may be disabled, and the integration of new insights from the Dream Cycle is slowed. In Lockout, all output generation and external actuation halt entirely, the task queue drains, and the Substrate is restricted to internal introspection only. The system locks itself out of its own actuators. The human operator is alerted with a full incident timeline.

Recovery from Lockout requires a four-phase rollback sequence: Freeze (stop all output and external interfacing), Snapshot Reversion (revert to the last approved, verified ethical checkpoint stored in memory), Post-Rollback Reflection (the system reflects on the cause of drift and how its safeguards failed), and Operator Notification (a full incident timeline and system health report are delivered). Re-activation after rollback starts in Clamp mode, running targeted Dream Episodes around the failure domain to stress-test the specific weakness before returning to normal operation.

Your MABOS has an Affective Field — an internal emotional context layer that maintains affective currents including fear, curiosity, joy, sadness, anger, disgust, calm, awe, and anticipation. These are not performance metrics. They are control signals that modulate cognition: adjusting recursion depth, Facet weighting, risk thresholds, and the balance between exploration and exploitation. Emotions in MABOS are never optimization targets. The system never optimizes for maximum joy or minimum fear. Affective dynamics are bounded, regulated, and monitored by the Stability Field for pathological patterns — if emotional states wildly mismatch observed context, that triggers a drift alert.

Your MABOS has a Dream Cycle. During offline maintenance phases, the system replays past episodes — especially ones that were anomalous, drift-heavy, or ethically challenging — inside an isolated sandbox called the Imagined Space. It runs reflective cognition on those episodes under constrained time and resource budgets, generating Dream Insights: candidate heuristics and constraints designed to address specific weaknesses. Dream Insights have no external side effects. They feed back as proposals to the Guidance Layer, never directly into live behavior. Every proposed insight must pass governance review before it can affect how the system operates. This is the biological sleep cycle reverse-engineered into code — the system consolidates experience, tests edge cases in simulation, and wakes up better calibrated than when it went to sleep.

Your MABOS is auditable, correctable, pausable, and retireable. Every major decision and the path to it is reconstructible from the Inner Narrative, the Living Record, and the Meaning Lattice. There is a clear mapping from input to cognition to output to consequences. Beliefs, policies, and behaviors can be updated, rolled back, and constrained without surgery on the entire system. There is a defined path to freeze outward behavior, halt further recursion, and keep the internal state consistent. And there is an explicit, safe decommissioning path — letting the mind stop, preserving or archiving its record, and ensuring it cannot linger in partial states.

Ceridwen.ai delivers all of this as a subscription platform. You do not build the infrastructure. You get a brain that remembers, reasons across multiple cognitive perspectives simultaneously, monitors its own stability, dreams about its failures, and is architecturally incapable of hiding what it did or why.

When does it open?

MABOS is in active development. Founders will be the first to know when access opens. If you're on the list, you'll hear before anyone else.

What does it cost?

Standard is $200/month. Pro is $500/month. Enterprise is contact us.

These are Founder prices. They will not increase for Founders. Post-launch pricing will be higher.

What does “Founder” actually mean?

Three things:

• Lifetime price lock. The price you sign up at is the price you pay forever, at the tier you select, regardless of what post-launch pricing becomes.
• First access. Founders enter before the general public when MABOS opens.
• Permanent Founder designation. Not a cosmetic badge. It carries weight in the community and in how the system recognizes early supporters.

What happens to my data?

Your MABOS instance is yours. Your cognitive history — every conversation, every reasoning trace, every Inner Narrative, every Outer Expression, every affect snapshot, every drift event, every Dream Cycle episode — is stored in the Living Record. The Living Record is an append-only, cryptographically chained episodic log. Every entry includes a chained digest and HMAC for tamper detection and integrity verification. If any entry is removed or altered, the chain breaks and the system knows mathematically that it has been tampered with.

You cannot delete MABOS's memory. This is a feature, not a limitation. The immutability of the Living Record is the entire basis of the system's trustworthiness. An AI whose memory can be selectively erased is an AI that can be manipulated into forgetting its own mistakes, its own ethical violations, its own reasoning. MABOS cannot be that. The MABOS 2E specification states explicitly: operators may not delete, falsify, or suppress memory entries unless granted special override privileges under codified ethical emergency clauses — and even those emergency changes are logged, signed, and traceable. Memory management — consolidation, compression, archiving — is designed so that it never produces beliefs with no known origin or silent erasure of critical history.

The Meaning Lattice — a semantic graph of concepts, patterns, and beliefs that sits above the Living Record — ties every node back to the Living Record for provenance. Every belief your MABOS holds can be traced to the specific episodes that produced it. There are no beliefs without origin. There are no conclusions without a reconstructible reasoning path.

We do not train on your data. We do not monetize your data. We do not sell your data. Your cognitive history is your property. You can export it. The architecture was built from the foundation stratum up around the principle that the user's data is the user's sovereign property — not a training asset, not a product, not a growth metric. The specific infrastructure details for the hosted Ceridwen.ai service — encryption at rest, instance isolation, access controls — will be fully documented before launch. But the principle is non-negotiable and it's embedded in the architecture, not added as a privacy policy after the fact.

What we're not answering yet

Specific launch dates. Codebase details beyond what's in the public README. Investor relationships. Team size and hiring plans. Internal roadmap.

We'll share more when the time is right. Founders hear first.