Browse documentation
Docs/Functionality

Grid Product Overview

Grid is a declarative computational language presented through a programmable workbook. It keeps the spreadsheet virtues people expect: cells, formulas, ranges, sheets, visible state, and…

productGrid 0.76.0Candidate docsCanonical contractVerified 2026-09-01

Grid Product Overview

Grid is a declarative computational language presented through a programmable workbook. It keeps the spreadsheet virtues people expect: cells, formulas, ranges, sheets, visible state, and the ability to begin calculating before every concept has a formal name. It adds semantic bindings, relational queries, reusable app-like surfaces, realtime inputs, geospatial tools, AI functions, Frames, tables, rules, simulation, optimization, exact symbolic mathematics, governed research workflows, and collaboration.

Grid is for teams that need a model to be both understandable and operational: finance workbooks fed by market data, geospatial models with map views, planning models with AI or trained-model scoring, operational dashboards with live connectors, and analysis systems where formulas are still the best authoring surface.

Models and workspaces

A model is Grid's code-and-data encapsulation: logic, schema, configuration and protocols written in Grid Language, together with its state and an optional interface. A workspace is the place where a person works with content, views and resources. Home is the workspace library: people create, find, open, rename, save and delete workspaces there without first having to learn what a model is.

Today each workspace in Home is backed by one model, and its name, folder and saved state are that model's. The surfaces below (Chat, Pages, Data, Code and the workbook surfaces) are views within a workspace over that same model. Code, the language reference and the APIs keep the model term, because they work with the model directly. The two terms are not interchangeable: a workspace is the surrounding place, and a model is the computational content inside it.

Product Promise

  • Spreadsheet-native authoring. Users write formulas and rules in a workbook mental model.
  • Spatial and semantic names. Models can begin with coordinates and ranges, then add names such as Revenue, prices, or risk_limit without abandoning the spatial calculation.
  • Declarative execution. Authors state values, relationships, queries, constraints, and reactions; Grid owns dependency order, incremental recalculation, execution strategy, and runtime placement.
  • Explainable qualitative reasoning. Natural facts such as design before launch and parcel inside district participate in bounded inference, reactive queries, contradiction detection, and WHY explanations.
  • Live computation. External inputs can stream into models through connector events or dense range ingress.
  • Multiple surfaces over one model. Sheets, datasets, tables, maps, charts, documents, and planning boards share one underlying model state.
  • Programmable functionality. Built-in functions cover spreadsheet math, text, dates, finance, statistics, geo, AI, frames, tables, and domain helpers.
  • Analysis and optimization built in. Beyond spreadsheet math, formulas can solve linear and mixed-integer programs, goal-seek and optimize with the SOLVE statement, integrate differential equations, fit curves, run spectral linear algebra and signal processing, and do exact symbolic mathematics—declare symbols, simplify and differentiate, solve equations under stated assumptions, compute bounded polynomial decompositions and Groebner bases, solve admitted polynomial systems, perform exact matrix algebra, and prepare certified integration and ODE work. Native FLINT-backed methods are available only in a Pro package that advertises the exact qualified capability; Grid-owned exact methods remain independent of FLINT.
  • Academic research as a governed workflow. The Pro Research Workspace seals current model evidence, applies immutable purpose-bound policies, manages bounded bibliographies and standards, prepares specialist-validation artifacts, and composes citable scholarly releases without turning Grid into a repository, DOI mint, HPC scheduler, or arbitrary-code runner.
  • Structured learning with a work record. Instructors can turn a protected model into an assignment, give every student a separate attempt, choose automatic or rubric checks, review the ordered answer history, and manage accommodations, appeals, answer-key release, and retention. Where deployment write policy admits the attempt model, students keep working in ordinary declared input cells rather than a separate quiz engine; the principal-bound learning answer route remains the owning write contract.
  • Desktop web-page work inside a model. A Browser surface can open an HTTP(S) page in an isolated Grid Desktop session, inspect bounded page and form state, perform deliberate interactions, and capture selected state for formulas. It does not share Grid sign-in state or silently fall back to an iframe in the web client.
  • Inspect the running product inventory. The capability browser identifies the running product and lets evaluators search availability, lifecycle boundaries, source families, evidence families, and explicit limitations without turning catalog metadata into permission or release authority.
  • Resilient coordination when selected. Coordination v2 provides exact-node mutual-TLS transport, replay protection, durable receipt/outbound recovery, and optional fenced DDIL authority for explicitly configured deployments. Its distributed lane is startup-selected and requires deployment-owned trust history and receiver idempotency; ordinary work does not pay its cost.
  • Operate the model directly. Calendar and Kanban let people drag, resize, and edit in place, writing changes straight back to the cells they are bound to; the App builder and Diagram add drag-and-drop composition of surface content.
  • Collaborative by default. The same model can be edited, viewed, and reused across sheets and richer workbook surfaces.

The Short Mental Model

  1. A workspace is where the user works; Home lists them, and each is currently backed by one model.
  2. A workbook is what the user sees inside it: sheets plus surfaces.
  3. A model is the executable state behind the workbook: reactive bindings, formulas, inputs, resident values, connector bindings, and metadata.
  4. Bindings can be addressed spatially (A1) or semantically (Revenue).
  5. Formulas and queries derive values; predicates describe what is true about or between symbols; rules react to values or schedules.
  6. Connectors bring outside data into the model.
  7. Frames handle columnar analytics; tables are the model-local record path.
  8. Geo adds spatial values, indexes, map views, and spatial kernels.
  9. Workbook surfaces turn the same model into maps, charts, boards, documents, datasets, and apps.
  10. Verified Learning records structured student work around admitted model inputs.
  11. Research Workspace governs evidence, policy, standards, specialist validation, and scholarly release over the same model and artifact store.
  12. Coordination services manage durable jobs, model routing, connector delivery, and—when explicitly selected—authenticated cross-node state movement.

Where To Go Next