Humans steward truth.
AI watches for signals.
Canon Engine is built on a conviction: knowledge for agentic systems must be governed, not just stored. The difference between a knowledge base and a knowledge platform is that the platform knows what's true, what's superseded, and what's in conflict.
Non-negotiable constraints.
Not guidelines — invariants.
These six rules constrain every implementation of Canon Engine. They cannot be relaxed, overridden, or deferred. They are the boundary between a governed knowledge platform and a vector database with extra steps.
Human authorship lineage
No artifact becomes Canon without human authorship lineage. AI can draft, humans steward. This isn't a philosophical position — it's an enforcement mechanism. Automated systems may propose documents for review, but promotion beyond draft requires a human steward's explicit decision.
Multiple signal requirement
No Canon entry originates from a single artifact observation. Pattern recognition requires multiple signals. A single meeting note doesn't become policy. A single email doesn't become architecture. Canon entries emerge from convergence — when the same truth surfaces repeatedly across different contexts.
Account + intent binding
Every artifact serves a purpose for someone. No knowledge exists in the system without being bound to an account and an intent. This eliminates orphan knowledge — the documents that exist because someone thought they might be useful someday but serve no identified purpose.
Canon domain mapping
No template hydrates without Canon domain mapping. Templates are governed by Canon schemas. When you generate a deliverable, a report, or an analysis, the terms it uses, the structure it follows, and the claims it makes are all traceable to Canon governance entries.
Conflict blocks execution
No execution derives instruction from an artifact in Canon-Conflict state. When two pieces of governed knowledge contradict each other, the system stops. It doesn't pick the newer one, the longer one, or the one with more embeddings. It surfaces the conflict for human resolution.
No direct storage writes
Consumers do not write directly to Canon Engine storage. All knowledge enters through the ingestion interface or the vault file system. This ensures enrichment is always applied, chunks are always generated correctly, and idempotency guarantees are maintained.
Canon is not a suggestion.
Canon is the authoritative body of system doctrine. A Canon artifact is a binding declaration that defines what the system is allowed to be, do, or assume.
Canon text establishes truth, authority boundaries, and non-negotiable constraints. Canon text is final unless explicitly superseded by new Canon. This is not a wiki. It is not a knowledge base. It is law.
Not a buzzword.
A formal discipline.
Intent engineering is the practice of ensuring every artifact in your knowledge system carries explicit intent — who it serves, what purpose it fills, and how it connects to the decisions your organization has made.
When your agents retrieve knowledge, they don't just get text. They get context: the status (is this accepted or draft?), the lineage (who wrote it and why?), the scope (which account does this serve?), and the confidence (how certain is the organization about this claim?).
This transforms retrieval-augmented generation from “find similar text” into “find authoritative, contextually relevant knowledge that your organization has explicitly endorsed.”
The difference is the difference between an agent that confidently cites a draft someone abandoned three months ago and an agent that surfaces the accepted policy your team debated and approved.
See how it's built.
The philosophy defines what Canon Engine believes. The architecture defines how it works.