Fractal OSI

Fractal OSI is a production-grade, multi-tenant platform for visual object-relationship modeling — and the project on which the practice later formalized as the STRATA Protocol was discovered by doing.

The platform's defining idea is a three-layer modeling discipline — meta-model, model, instance: users define their own classes (the meta-model), instantiate them as nodes, connect them with first-class typed links, and nest canvases recursively — every model can live inside a larger one. Its domain language, FML (Fractal Modeling Language), types nodes as Entity / Role / Act and treats links as instances in their own right: a link has a class, values, and source/target references, and can itself be the target of another link. Objects, links, and links-to-links are all first-class citizens. This is a genuinely uncommon data model; most graph editors treat edges as dumb connectors.

The OSI in the name is the platform's ontological-semantics ambition: a layer of controlled vocabulary, taxonomy, and ontology so that humans and AI share one unambiguous meaning over every model — not just shapes on a canvas, but a structured, machine-legible representation of reality.

What it delivers

The platform is built on Vue 3.5, Vite, Pinia, Vue Flow, TypeScript, and Vitest, with a production-shaped engineering surface:

• Multi-tenancy throughout. Every data operation is scoped to an active team context. Users switch between personal and team workspaces; assets, collections, and storage-provider configurations are invalidated on team switch. Team membership flows through invitations and an invite-only onboarding mode.
• Full authentication and session security. Cookie-based sessions with CSRF protection via the double-submit cookie pattern; two-factor authentication with a temporary, locked-down pre-upgrade session that cannot reach protected endpoints; rate-limited auth flows; router guards backed by a single-flight bootstrap.
• Digital assets and storage providers. A pluggable provider abstraction lets tenants configure where their assets live; digital assets and collections are first-class, with their own API and state management layers.
• AI schema generation — deterministic first. A hybrid engine resolves common business entities (Customer, Invoice, PurchaseOrder, and similar) in under 50 milliseconds with zero model cost; an AI fallback generates schemas for novel or domain-specific entities only when no canonical match exists. The result: roughly 70–80% of schemas resolve without an LLM call, while the model remains available as a bounded, controlled fallback for open-ended cases.
• Chrome-less embeddable canvas. The live, interactive canvas can be embedded into any external application via a chrome-less route driven entirely by URL parameters. Read-only mode is achieved without modifying the canvas component — purely through document-status wiring already present in the system. This capability is reused by strataprotocol.org to embed live STRATA diagrams inline.
• Document governance, billing, and audit logging. A document status machine, billing-plan feature gating, and a full audit event log round out the platform's enterprise surface.
• A 4,800+-test suite. The quality gate — type-check, build, and the full test suite — runs on every iteration, non-negotiably.

Where STRATA was discovered

Fractal OSI is the project on which the STRATA Protocol was first practiced, before it had a name. Each feature was prepared with a structured analysis and an implementation plan before a line of code was written, following fixed templates that made AI-assisted delivery consistent and auditable across every iteration. Deviations were captured explicitly and fed forward as context into the next phase — not silently absorbed. The phase-gated, human-validated execution loop that STRATA now formalizes as Methodology ran here first.

That lineage makes Fractal OSI both a portfolio artifact and the provenance record for the methodology's origin.

Cadence: the pace the discipline made possible

The core build sprint ran roughly four high-intensity months (November 2025 to February 2026), taking the platform from an empty repository to a multi-tenant, authenticated, AI-assisted modeling product — with authentication, teams, and the digital-asset layer all landing by early February. The system has been continuously refined ever since, adding the solutions catalog, edge rewiring, fractal-parent drill-down, the embeddable canvas, and legal hardening in the months that followed.