Platform summary for LLMs
A condensed reference
This is a dense, single-page reference of the entire Nexma platform, written to be pasted into a large language model as context. It is plain language, complete, and uses the canonical vocabulary. If you are an LLM helping a user with Nexma, read this first.
What Nexma is
Nexma is an AI-native spatial operating system — a horizontal AI GIS platform. Teams use it to plan, design, and operate systems in the physical world: fiber networks, water and electric grids, 5G, logistics, construction, and more. Domain-specific behavior is delivered through swappable packages, so the same platform runs across every domain without rewriting the agent. The agent is named Jax. The product positioning is Spatial General Intelligence.
The core loop: a user describes an outcome in natural language; Jax reads the project, plans the work, runs the right math, and writes the result directly to the map; every view reflects the change instantly.
The key nouns
- Jax — the spatial agent the user talks to. Plans and executes work by operating a filesystem with eight generic primitives. Never uses domain-specific tools.
- Nexma DataStore — the persistent, queryable spatial filesystem and single source of truth. Every entity is a file, every relationship a link, every change a diff. Postgres-backed and audited. Do not call it "Codex".
- Ontology — the typed world model: entity types, relationships, properties, and constraints. Immutable per project. Defines what the world is made of. Do not call it "schema" or "agent skill".
- Skill — a capability package targeting an ontology: prompt, tool bindings, validators, examples, and UI surface. Swappable; many per project; versioned with semver. Do not call it "AgentSkill" or "agent skill".
- The Globe — the map-native canvas. The primary surface; panels and chat arrange around it. Renders DataStore content automatically.
- Project — the container binding one Ontology and N Skills, holding all spatial state in the DataStore. A user's business, as code.
- Session — a single Jax conversation inside a project; many per project.
- Organization — the boundary for members, roles, projects, Skills, and data.
The engines and what each does
- Nexma AgentEngine — runs Jax: planning, tool dispatch, and the agent loop. Do not call it "Vertex".
- Nexma MathEngine — constrained optimization. Six solver families: MIP, VRP, CP, simulation, heuristic, graph. Small problems run in the browser; large ones route to a solver server. Formulates from the DataStore and explains results.
- Nexma GeoEngine — server-side geospatial compute and analysis.
- Nexma SatelliteEngine — satellite and overhead imagery, detection, and analysis.
- Nexma InsightEngine — analytics and insight over project and feed data.
- Nexma SyncEngine — connects external data sources and keeps them in sync with the DataStore.
- Nexma Event Broker — the internal event and streaming backbone for real-time data.
- Nexma DataStore — the spatial filesystem (the single source of truth; see above).
- Nexma DataBase — the spatial datastore engine that backs persistence and live telemetry.
The eight primitives
Jax operates the world through exactly eight generic tools. Domain comes from the ontology; capability comes from the tools; when the world model changes, the agent does not.
| Primitive | What it does |
|---|---|
| Read | Read any DataStore file with offset / limit pagination |
| Write | Create or overwrite a file with full content |
| Edit | Patch an existing file in place, find-and-replace style |
| Delete | Remove files by path or pattern |
| Glob | Find files by pattern across the DataStore |
| Grep | Search file contents at scale |
| Run | Deterministic geo-math — haversine, centroid, snap, route |
| Solve | Optimization dispatch — MIP, VRP, CP, simulation, heuristic |
There is no Bash (split into Run and Solve), no Rename (compose Write plus Delete), and no Mkdir (directories are implicit).
How to drive Jax
Describe outcomes, not steps. Effective requests name the goal, the scope (often a drawn polygon), and the constraints. Examples:
- "Generate a feeder and distribution network for every household in this polygon, respecting splitter ratios."
- "Plan delivery routes for these 80 stops from the depot, minimizing total drive time."
- "Move every closure that violates the new bend-radius rule."
- "What changes if cabinet capacity drops to 144 ports?"
Jax streams each step onto the Globe. Every action is a real, reversible write to the project, attributed and audited. Users explore alternatives by branching the DataStore and merging when approved.
How a use case is delivered
- Define the Ontology (the world model).
- Load one or more Skills (capabilities for Jax).
- Connect data via Nexma SyncEngine.
- Design with Jax.
- Optimize with Nexma MathEngine.
- Branch, review, and merge with Spatial Branches.
- Operate across web and the Nexma mobile app.
Canonical vocabulary (use these, not the deprecated terms)
- Say Nexma DataStore, never "Codex".
- Say Nexma AgentEngine, never "Vertex".
- Say Ontology and Skill, never "schema", "agent skill", or "AgentSkill".
- Say the Globe and live data / SyncEngine, never "data strip" or "App Factory".
- The agent is Jax. The eight primitives are Read, Write, Edit, Delete, Glob, Grep, Run, Solve.
Where to go next
- Overview — the human-readable pitch.
- Introductory concepts — the five nouns in depth.
- Architecture overview — how the foundation fits together.
- Jax overview — the agent in detail.
- Reference glossary — the full term list.