Sovereign Cognition
The model serves.Indwel governs.
Indwel places probabilistic intelligence inside a governed system. Objective, Evidence, Memory, authority, effects, and Settlement remain outside the model. Use frontier, private, or local inference without making the model the operating system.
See the whole cognitive actThe whole cognitive act
The output may be probabilistic. The business result is governed.
A Cognitive Contract bounds the act before inference begins: Objective, permitted information, methods, human authority, effects, completion conditions, and required proof.
The system keeps the offices separate: context from Evidence, recommendation from Decision, permission from Effect, attempt from Observation, and generation from Settlement.
- EvidenceAdmitted facts and sources
- CognitionReason, retrieve, compare, calculate, ask
- AuthorityHuman or institutional Decision
- EffectOnly the permitted external action
- ObservationWhat actually happened
- SettlementWhat becomes durable truth of the Work
Cumulative cognition
Indwel can remember vastly more while asking the model to carry vastly less.
Work persists as governed state rather than conversational history. A Work Brief compiles the smallest constitutionally sufficient state for the next cognitive act.
MemoryUser-governed continuityNot Evidence
Cognitive CapitalDecisions, assumptions, questions, risks, methods, obligations, boundariesThinking made durable
ContinuityLineage across sessions, people, and timeNot transcript replay
Library PacksAdmitted informational worlds for bounded cognitionSource identity preserved
Temporal CognitionFreshness, deferred Work, scheduled review, future conditionsTime can change what is known
The smallest constitutionally sufficient state for the next act
Cognition under separation of powers
Capability does not confer sovereignty.
Indwel can interrogate, retrieve, reason, plan, coordinate, use tools, and invoke specialist agents. Each capability receives a bounded role under the Work; none inherits authority because it can act.
Prepared Reciprocity asks only where a human contribution can materially change authorized Work—and shows what is known, what remains unresolved, why the answer matters, and what changes next.
Inference freedom
Choose the model. Keep the authority.
The Inference Firewall makes model choice an enterprise capability rather than an architectural surrender. Change providers, combine models, route specialist work, or bring inference entirely inside your own boundary without moving the sovereignty of the Work.
Objective · Evidence · Memory · authority · effects · continuity · Settlement remain outside model sovereignty.
Amazon Bedrock
Broad managed model choice behind the same Indwel authority boundary.
OpenAI · Anthropic · Google Gemini · Vertex AI
Plus Azure OpenAI, Microsoft Foundry, Mistral AI, Cohere, xAI, and OpenAI-compatible endpoints.
Private VPC · local runtimes · customer-hosted endpoints
Keep model execution inside the operating boundary your institution requires.
Air-gapped inference
Operate inference without a public-network model dependency while preserving the same Work constitution.
One Objective, Evidence authority, action authority, and Settlement spine remain intact.
The governing relationship
Your models work for you—not the other way around.
Indwel puts frontier, private, or air-gapped intelligence inside one governed operating system without handing models ownership of Evidence, Memory, authority, action, or terminal truth.