Poesis
Partnership work: implement KnowledgeSources, model integration, client reduction, provenance transfer, and Definition Manager registration.
LLM vendors · Partnership
LLM Vendor Partnership — Value Proposal
Test what explicit context changes. This partnership proposes comparisons between governed, attributable context and document-retrieval baselines on selected tasks. GSM content, SIE foundations and local SAF controls support that experiment; model-facing integration, operational assurance and measured uplift remain work.
How to read this proposal. Present tense describes the Poesis reference design and target operating model which might not yet be fully implemented or operational. Partnership adherences described below may therefore include work to advance Poesis components from their explicitly stated maturity levels.
The partnership thesis
Five reasons to partner.
One partnership addresses two limits: structured context raises the capability ceiling; enterprise platform, market access, platform advantage, and sovereignty raise the deployment ceiling.
Capability ceiling · Structured context
Better intelligence without bigger models.
A model that trains and reasons against explicit, causal, governed, and snapshot Definitions of the world works against explicit structure rather than relying only on correlation. That structure provides a substrate for impact analysis, counterfactual testing, and planning against modeled dependencies and much more...
Supported by adherences
Don't compete on larger models. Compete on better intelligence.
Deployment ceiling · Enterprise platform
From language model to enterprise AI platform.
Companies buy solutions, not models. Partner models provide the intelligence; Poesis provides structured organizational context, human loop, deterministic evaluation of inference, grounded enterprise agentic development workflows, and audit evidence for accountability. Together they form a deployable enterprise agentic platform — able to act, not only answer.
Supported by adherences
Transform a model into an enterprise AI platform.
Deployment ceiling · Market access
A credible path into regulated industries.
Enterprises don't stall because AI isn't powerful enough — they stall because they can't trust it to act. Explainability, governance, auditability, and policy enforcement are entry conditions, and Poesis is designed to provide their shared context and control substrate.
Supported by adherences
Governance is the entry ticket to regulated industries.
Deployment ceiling · Platform advantage
A defensible advantage beyond model performance.
Grounded inference, governed execution, and model specialisation create a combined platform that is harder to commoditize than standalone inference. It moves the commercial relationship into AI governance, digital operating models, enterprise transformation, and regulated automation — increasing deal scope and strategic relevance.
Supported by adherences
Build a platform advantage, not simply a model advantage.
Deployment ceiling · Sovereignty
The foundation of a sovereign AI stack.
Not "a sovereign LLM" — a sovereign AI stack. You provide the sovereign intelligence. Poesis provides the governed context. Together they form the trusted infrastructure enterprises need to deploy AI at scale. As a neutral, model-independent substrate, GSM/SIE could be advanced as a candidate foundation for future European standardization of governed enterprise AI.
Supported by adherences
Build the sovereign AI stack, not only a sovereign model.
The value mechanisms / adherences
How the partnership creates value.
An adherence is a concrete interface where partner capabilities and Poesis components combine to create reciprocal value. Each business outcome is supported by one or more such mechanisms.
The value flows
An interactive map of the exchange.
Explore the map. Click a flow to see its mutual value. Click an organisation, LLM, or Poesis component to see every adherence it holds. Expand a Poesis box for internal detail. Solid wires are partnership adherences; dotted orange wires are the organisation’s own governance loop.
Core invariant: models PROPOSE (inferred, confidence < 1.0) · accountable humans dispose through ITIP · SIE evaluates constraints and executes governed Mechanisms — PROPOSE / DISPOSE.
Organisation operations
the customer organisation retains supervision, regulation, and governance over its own governed truth — that authority is never delegated to a model, and Poesis is the instrument through which it is exercised
No direct adherence exists between the organisation and the models. The organisation reaches them only through the governed perimeter — which is what makes supervision, regulation, and governance enforceable rather than declarative.
LLM operations
vendor-neutral by design — any sovereign or frontier model: self-hosted, sovereign-hosted, or open-weight
Poesis stack
target topology: organization-scoped instances; isolation, egress and recovery require acceptance; federation is future work
sie-definition-manager — hosts GSM, Ascription lifecycle, generative governance grammar
sie-definition-blackboard-manager — contribution registry and Blackboard-seal lifecycle
sie-operator — executes in-effect Mechanisms and validates typed inputs and effects; automatic Norm preflight remains to be built
itip-web-application — planned human gate for intent, approval, simulation, appraisal, and audit
gsm-ontology — selected GDPR/NIS2 templates and TOGAF/ISO vocabulary; broader frameworks and operational binding remain planned
itip-definition-blackboard-*-sourcer — designed to wrap LLMs / agents as KnowledgeSources over code repositories; other source classes require adapters
orchestrator-agent — portfolio / program / team
specialist-agent — architecture, dev, QA, security, docs agents
harness — resolution, injection, routing, validation, agent authorization and journals exist in source; human authority, transaction and host compatibility need acceptance
Maturity disclosure
Operational maturity by value flow.
Current maturity and partnership work are stated separately for each flow. Implementation is part of the exchange: a partner can contribute model access, engineering, evaluation, and applied research to turn selected flows into measured operational proofs.
Partnership work: build the model-facing context and tool surface, then measure it against document-retrieval baselines.
Partnership work: assemble and validate the ITIP human gate, Norm and provenance evaluation, disposition audit, and regulated-estate proof.
Partnership work: agree corpus permissions and oracle criteria, run specialisation experiments, and measure uplift.
Partnership work: define model evaluation contracts, datasets, baselines, trace capture, assurance criteria, and a repeatable service surface.
Partnership work: qualify existing resolver and supported host bindings; complete workload evidence, model-version governance and outcome-based selection.
Partnership work: accept parser/contracts, authenticated decisions, atomic writes and supported host behavior; collect comparable outcome evidence.
Cross-cutting operational adherence · Co-implementation
Direct value for Poesis: faster implementation, stronger experiments, benchmark evidence, and production feedback across the reference design.
Indirect value for the LLM vendor: a governed enterprise integration proving where its models perform, fail, and improve under accountable operating constraints.