Understand
A typed graph for product knowledge.
Product teams already think in entities and relationships: personas, jobs, opportunities, features, outcomes. UPG names them, types them, and connects them so the graph can be queried, validated, and shared across tools.
“A product team writes 28 documents and 150,000 words per product. That’s a lot of unstructured prose for an AI to parse, a system to query, or a team to keep consistent. A graph stores the same knowledge as typed entities and explicit connections: less prose, more structure, clearer answers.”
28 documents. 150,000 words. Zero structure.
Every product team writes the same artefacts: PRDs, OKRs, roadmaps, personas, Jobs-to-be-Done docs, Business Model Canvases, RICE sheets, retros. Each lives in its own tool, in prose, with its own vocabulary.
The same entity (a persona, a feature, a job) appears in a dozen places under a dozen names. No system can answer “which features serve which jobs for which persona?” without a human stitching the prose together by hand.
AI inherits the same problem. Prose is hard to parse reliably. Prose drifts. Prose contradicts itself.
Typed nouns. Typed verbs. One graph.
Replace each artefact with a small set of typed entities and explicit edges. A persona becomes a persona node. A job becomes a job node. The fact that the persona pursues the job becomes a persona_pursues_job edge.
The graph carries the same knowledge as the prose, with three differences: every concept has a stable type, every connection is queryable, and every claim can be validated.
AI can read it. SQL-like queries can traverse it. Two teams looking at the same graph see the same product.
Four primitives. The whole spec composes from these.
Nouns
Entities
Typed nodes: persona, job, opportunity, hypothesis, feature, metric. Each carries a stable id, properties, and a maturity tier. Same vocabulary across every team and tool.
Browse entity types →Verbs
Edges
Typed relationships: persona_pursues_job, feature_addresses_job, hypothesis_targets_outcome. Edges carry direction, semantics, and validation rules.
See edge catalog →Geography
Regions
Coherent neighbourhoods of entity types: Discovery, Strategy, Execution, Validation. Each region has a mental model, an anchor type, and the atomic domains it composes.
Tour the regions →Projections
Frameworks
Business Model Canvas, Lean Canvas, Opportunity Solution Tree, RICE, Jobs-to-be-Done: each is a view onto the same graph. Pick the lens you need without rewriting the underlying knowledge.
See frameworks →UPG doesn’t replace what works. It unifies it.
Business Model Canvas, Lean Canvas, Opportunity Solution Tree, RICE, Jobs-to-be-Done, Wardley Mapping. Every classic framework is a projection of the same underlying graph. Pick the view you need for the conversation you’re in.
The work you already do, the artefacts you already produce, is the graph. UPG just gives it a stable shape.
Specification
Every entity type, every edge, every validation rule. Versioned, MIT-licensed, open.
Read the spec →Why it exists
The argument for replacing 150,000 words of product prose with a typed, queryable graph.
Read the manifesto →See it
Walk the catalog interactively, every type, every edge, every property, with live cross-links.
Open the explorer →Use it
Business Model Canvas, OST, RICE, Lean Canvas, Jobs-to-be-Done, Wardley. All mapped to UPG entity types.
See frameworks →