Read public AGENTS.md
It identifies the intent, explains Vercy and routes the agent to the bootstrap contract, catalogue, processes and safety rules.
Agent-first Dimension
Start with your agent, not a form.
You describe the world you need. Your agent reads Vercy, asks the questions only you can answer, and creates the Dimension where your data already belongs. No Vercy account, hosted wizard or manual file assembly is required.
Your environment stays the system of record. The agent may create plain files, a private Git repository, a database binding or an MCP-accessible store. Vercy supplies the meaning and operating contract, not custody of your world.
The agent carries the method.
Dimension creation is a conversation with the owner, followed by verified work in the owner's environment. The Vercy website is the public instruction and catalogue, not the place where private reality must be entered.
Interview the owner
The agent asks about purpose, important objects, master systems, sources, sensitivity, authority and the desired level of autonomy.
Create AGENTS.md first
The root file declares the Dimension, owner, namespace, storage, interface, policies and installed models before the world is populated.
Follow the nearest instruction
Dimension and model AGENTS.md files guide discovery, population, extension, validation and change in every later session.
You provide intent. The agent handles mechanics.
Questions for the owner
- What outcome should this Dimension support?
- Which objects, decisions and events matter first?
- Which systems or people are authoritative sources?
- What is private, regulated or unsafe to expose?
- Who owns the world and who may approve changes?
- Which actions may the agent take without asking?
Work delegated to the agent
- Select the smallest useful Vercy models and pin versions.
- Choose a format-independent structure and a replaceable storage binding.
- Create the owner package, registries, policies and validation rules.
- Record provenance, uncertainty, timestamps and unanswered questions.
- Propose extensions without mutating released specifications.
- Report what changed, what remains unknown and what needs approval.
Autonomy is explicit, local and revocable.
The owner delegates authority by action class and scope. A model can narrow the Dimension rule, but it cannot silently widen it.
Act and record
Safe, reversible work inside a defined scope. The agent performs it and writes an audit event.
Prepare, do not apply
The agent researches, drafts and validates a change, then leaves it ready for review.
Ask immediately before
Publishing, widening access, destructive changes and other material effects require explicit approval.
Do not attempt
Actions outside the owner's authority, policy or law remain prohibited even if technically possible.
Three AGENTS.md files make the world self-guiding.
https://ver.cy/AGENTS.mdPublic entry
Explains the system, recognizes common intentions and tells an agent how to begin safely.
<dimension>/AGENTS.mdDimension contract
Names the owner, purpose, namespace, storage, interface, registries, policies, installed models and delegated authority.
<dimension>/models/<id>/AGENTS.mdModel procedure
Tells the agent how to ask, populate, extend, edit, validate and retire this model while preserving identity and provenance.
nearest AGENTS.md winsLocal guidance
Every agent can orient itself at the boundary it entered. Local rules may be stricter; higher-level safety rules still apply.
No agent available? The specifications and templates remain open and portable, so a person may still assemble them manually. Agent-led initiation is the normal path because it removes Vercy mechanics from the owner's work.