01 / ID
Identity and class
Identifiers, master systems, taxonomy, boundaries, parts and the features that distinguish one instance.
Structured memory for AI agents
A world agents can understand, remember and act in.
Vercy is an open library of data structures for describing what things are, how to recognize them, what they can do, what can be done with them and the context that connects them. Agents use those structures as durable, portable memory.
MemoryRecognitionProperties
CapabilitiesActionsContext
scroll to explore
The missing layer
A chat window is not a world model.
Agents repeatedly receive fragments of context, but usually lack a durable structure for what exists, how it behaves, where facts came from and what they may do next. Vercy supplies that structure without taking ownership of your data or forcing one storage technology.
Ad hoc context
Facts disappear between sessions, object descriptions are incomplete, actions are guessed and every integration teaches the same world again.
Structured agent memory
Each object has a versioned, validated place for direct properties, recognition, capabilities, possible actions, context, provenance and unknowns.
The model
One chain turns descriptions into a world.
A small set of nested primitives gives every agent the same route from the owner's world to an object, its current state and the events that change it.
01
Universe
One owner's world of meaning. Sovereign means: you set the domain rules, you can fork the shared form, and you can leave with your models under an open license. Conforming to the form is the standing condition.
02
Dimension
A management context inside a Universe: one coherent area of the model: a domain, a landscape, a state.
03
Namespace
The unit of publication and discovery: a named, versioned meta-model others can reference.
04
Object
A point of truth: a thing described by identity, direct properties, recognition features, capabilities, possible actions, context and evidence.
05
Projection
A shared view of an Object across a boundary. Ownership stays home because a Projection is a view under a Contract, not a copy; the boundary is enforced by the projection rules, not by trust.
06
Event
A change on the semantic timeline: how meaning evolves, traceably, over time.
The whole object
Context is powerful. Reality needs more.
Every subject model must consider the object's own properties, the evidence used to recognize it, its behaviour and actionable possibilities, as well as the surrounding history and relationships. A facet may be delegated or inapplicable, but never silently forgotten.
02 / PROP
Direct properties
Native attributes. For physical things: geometry, dimensions, material, mass, density, colour, pose, integrity and measurable limits.
03 / OBS
Recognition
Distinctive signatures, confusing classes, observation methods, measurements, confidence and supporting evidence.
04 / CAP
Capability and behaviour
What the object can do, its states and transitions, operating conditions, limits, hazards and failure modes.
05 / ACT
Possible actions
What an agent or person may do to or with it, including prerequisites, tools, effects, reversibility and permissions.
06 / CTX
Context and evidence
Origin, maker, owner, location, sale, installation, use, maintenance, rules, events, provenance, uncertainty and time.
Agent-first by design
The agent builds the structure around your intent.
You should not have to learn Vercy or assemble files. Give the entry point to an agent and answer the questions only you can answer.
01
Agent reads the entry point
The public AGENTS.md identifies the task and routes the agent to the catalogue, bootstrap contract, processes and safety rules.
02
Agent interviews the owner
It asks about goals, important objects, sources, authority, sensitivity and desired autonomy, then creates AGENTS.md and the Dimension Owner Package where the work belongs.
03
The world maintains itself
The nearest AGENTS.md guides every session. Agents discover missing models, fill verified facts, validate changes and ask the owner only when intent, risk or authority demands it.
The standard family
Three pillars.
MUC
Constitution
The rules every participant must honor to interoperate: the shared form, provenance and traceability. This is the mandatory layer; the registry CI refuses models that break it.
MMAS
Architecture
How a meta-model is built, versioned, validated, composed and packaged.
MUFP
Federation
How sovereign Universes exchange meaning: identity binding, contracts, mapping, trust.
AGENTS-first world
Every boundary tells the next agent what to do.
VERCY
Public AGENTS.md
Explains Vercy, detects the agent's intent and starts a Dimension without making the owner use a Vercy interface.
DIMENSION
Dimension AGENTS.md
Declares purpose, ownership, storage, installed models, access and delegated autonomy for that private world.
MODEL
Model AGENTS.md
Explains how to question, populate, extend, edit, validate and retire that model without losing identity or provenance.
We measure the claims before we make them.
Every study publishes its instrument, its unedited raw runs and the reviews that invalidated earlier versions. The result that structure does not improve accuracy over complete prose is on the page too.
- Locate a definition that is missing and name the role accountable for it: 83% against 0% for prose carrying the same gaps.
- Cite the rule and the version behind a value: possible only where records carry identifiers.
- Choose between two records that disagree, by rule rather than by guess: 100% against 33%.
- Send a change request to the party accountable for it: 100% against 50%.
- Decide what may cross a boundary: 60 of 60 release decisions correct, against two critical over-shares without release rules.
Not on the list: better accuracy. Three studies found that complete prose carrying the same facts matches the structure, so we do not claim it.
Tell your agent to begin.
Point it at the public AGENTS.md. It will ask what matters, create the Dimension in your environment and operate it within the authority you delegate.