TECHNOLOGYTECHNOLOGY

Nexma DataStore

Nexma DataStore — the Codex — is a persistent, queryable filesystem for the physical world. Every entity Jax knows about — every closure, every transformer, every aircraft, every ship — is a file. Every relationship is a link. Every change is a diff.

Domain knowledge is decoupled from the agent. Jax reads and writes Codex paths through eight generic primitives; the schema lives in the Ontology, the operations live in Skills, the audit trail lives in Branches. One filesystem, every domain.

Persistence

Postgres-backed, audited

Access

Eight generic primitives — Read · Write · Edit · Delete · Glob · Grep · Run · Solve

Scope

Org · Project · Session

The Ontology

An Ontology defines what the world is made of. Entities, relationships, properties, and the constraints that bind them. Load FTTH — Jax designs fiber. Load water — Jax designs water. Same agent, different world.

Ontologies are typed, versioned, and swappable per project. Constraints are first-class — splice ratios, voltage drop, pressure budgets, regulatory geometry — and validated on every write. The agent never hardcodes a domain; the domain is the Ontology.

Form

Typed graph — entities, links, properties, constraints

Authoring

ReactFlow canvas · natural-language maker

Distribution

Templates per vertical — fork, extend, govern

The Skill System

A Skill is a bundle of capability targeted at an Ontology. Prompt, tool bindings, validators, examples, and the cinematic UI surface that goes with them. One project binds one Ontology and any number of Skills.

Skills are versioned the way code is versioned — semver, with major bumps when capabilities are removed. They compose: Aerial Investigation, Cyber Threat, SIGINT Geoposition, FTTH Network Design — each one a swappable expansion of the same agent.

Authoring

Sectioned editor — identity, prompt, tools, validators, examples

Versioning

Semver — major / minor / patch

Distribution

Org library · marketplace-ready

Spatial Branches

Git for the physical world. Branch the Codex to explore an alternative — a different cabinet layout, a different patrol route, a different incident scenario — without disturbing the truth. Merge the branch back when the plan is approved.

Every spatial decision is reversible. Every alternative is durable. Reviewers see exactly what changed — entities added, relationships rewired, constraints violated — the same way a code reviewer sees a pull request. Main is the operational truth; branches are the imagination.

Model

Main · feature branches · merge requests

Granularity

Entity-level diffs with constraint validation

Audit

Full history per branch, per file, per author

The Math Engine

Spatial intelligence is constrained optimization. The Math Engine is the solver layer Jax dispatches to: mixed-integer programming, vehicle routing, constraint programming, simulation, heuristic search — six families, one interface.

Small problems run in the browser via GLPK and bespoke heuristics. Large problems route to the Python solver server on Railway. Jax chooses the family, formulates the problem from the Codex, and explains the answer in plain language.

Families

MIP · VRP · CP · Simulation · Heuristic · Graph

Dispatch

Browser ↔ Solver Server — automatic

Explainability

Self-correcting, citation-traced