The Model Workspace

Model, Analyze & Solve is the studio where an optimization model is the protagonist: you build it, analyze it, solve it, and improve it step by step — all in one workspace, without hopping between disconnected screens. Open it from the sidebar (Model, Analyze & Solve) or go straight to /studio.

Everything in the workspace is a view of one canonical model. Whichever way you author it — visually, by chatting with the AI, or as JSON text — you are editing the same model, and the other views update to match.

Creating a model

From My Models, choose New Model to open the launcher. Every starting point creates a first-class model you can version and reuse:

Blank — start from an empty model.
Describe with AI — write what you want to optimize and let the assistant draft it.
Visual — lay the model out on the canvas.
Editor — write the model as JSON text.
From a template — pick one of the 102 curated templates; it is materialized into a new model.
From the marketplace — open a published model with Use in studio.
Import a file — bring in an MPS, LP, CIP, or JSON model.

Templates, marketplace models, and imports all seed a real model that lands in the workspace — born with a first version — instead of a one-off solve you lose afterwards.

Build · Data · Analyze · Solve

The workspace has four top-level tabs, and a model-at-a-glance rail (variables by type, constraints, density, problem class, health) that stays visible on every tab.

Build

The Build tab holds four authoring lenses for the same model:

Canvas — a visual graph of variables, constraints, and the objective. (Very large models skip the canvas automatically and are edited from the other lenses.)
Assistant — the AI builder, scoped to this model. Describe the model in natural language, then keep chatting to refine it ("add a variable y", "make x integer", "tighten the budget to 100"). Each reply updates the same model and autosaves. You can attach a spec or CSV to ground the formulation.
Editor — the model as JSON text. A valid edit reflects on the canvas and autosaves; malformed JSON shows an inline error and blocks solving until it is fixed, so you never act on a model you only think you changed.
JModel — a declarative modeling language: sets, params, and indexed families that compile into the flat model. The lens that unlocks datasets and scenarios. (Instance-gated.)

Data

The Data tab manages the model's datasets — named data bundles that fill the sets and params a JModel formulation declares, so one model can run many scenarios.

Analyze

The Analyze tab summarizes the model — objective sense, a constraint operator breakdown, terms per constraint, problem class — and hosts Explain this model: a grounded, plain-language account of what the model optimizes and what stands out about its structure. The numbers are computed in Python; the AI only narrates them, so it never invents figures.

Solve

Pick a solver (or Auto, which routes by problem class) and run. Solving streams in real time as Live Solve: a convergence chart of the primal incumbent against the dual bound (the gap closing), plus gap %, nodes, incumbents, and elapsed time. When it finishes, the panel shows the result and which solver actually ran.

A running solve is durable: it lives on the server, not in your browser, so it survives switching tabs, reloading, duplicating the tab, moving to another device, or losing power. Reopen the model and a still-running solve re-attaches; a finished one shows as the last run.

Versioning

The workspace separates two kinds of saving:

Autosave keeps your draft safe continuously as you edit — silent checkpoints.
Commit records an explicit, message-bearing version: a milestone with a required "What changed?" and an optional "Why?". A suggested message is derived from the diff.

From the version history you can review past versions, see the diff between any two, restore an earlier one, and ask the AI to explain the changes between two committed versions. Versions are immutable snapshots, so a model carries its own history of how — and why — it evolved. See Versioning for the full picture.

Publishing

When a model is worth sharing, publish it to the marketplace from the Analyze tab's Marketplace card — publishing pins your latest committed version to a public listing that other teams can fork. See Publishing Models.

Running history

Each model keeps its own run history, and the global executions page lists every solve across the platform with its origin, model name, and result — so a studio run is always traceable back to the model it came from.