How it works
The protocol uses a small set of building blocks to organize different kinds of work. Parts of it are still being built.
The ecosystem verbs
The protocol describes five actions:
Connect. A person, organization, or AI joins with a record of preferences, risk appetite, cash-versus-equity ratio, and operating constraints. The model also draws on vocabularies such as Human Design, numerology, and astrology to describe preferences.
Run. A Formation works toward a goal. A Run records who took part, their configurations, what was produced, and associated metadata.
Template. A successful Run can become a reusable pattern for future Formations.
Fund. Work can be supported with cash, equity, in-kind capacity, or equity across organizations. Funding and remuneration are part of the protocol's design.
Graduate. A Formation, a sub-unit, or a person moves into a new arrangement: spinning out as an entity, joining a portfolio company, or handing work off to a template.
The Resolution Mesh
A Formation is composed out of six kinds of nodes:
- PureAI — an AI agent operating against a defined scope, no human in the loop for the routine path.
- Reference — a static or slowly-changing body of material (documents, datasets, prior Runs) that other nodes consult.
- HumanFrontedByAI — an AI that mediates a human's contribution: drafting, scheduling, triaging, so the human shows up only where human judgment is actually needed.
- PureHuman — a person operating directly, no AI intermediation. Used where the task requires it.
- Organizational — a whole org as a node: another company, a partner team, a portfolio venture acting as a single compositional unit.
- Bridge — a node whose job is translation across contexts — between humans and AI, between organizations, between domains.
The mesh combines these different kinds of participants. Its design uses Celery for orchestration and a Command → Agent → Skill pattern to express behavior.
AppAlchemy and Folding
AppAlchemy is the practice of composing products from Resolution Mesh graphs. In this approach, an application is an arrangement of nodes with an interface, and existing Runs can inform new applications.
Folding extracts a reusable pattern from a Run, making it available as a template or input to another Formation.
Learning from Runs
The Runs collection records how people and AI worked together, what they produced, and metadata about the inputs, process, and outputs. We value it as a way to learn which combinations work for different problems, constraints, and costs.
The protocol is designed to collect and fold that information across engagements. Its validation-first approach draws on Karpathy's autoresearch pattern.
