Public product preview

Optimization that stays live.

Describe the problem, attach public data, and choose an exact or evolutionary route. MAPLE Harness builds a checked optimization workspace, shows numerical progress, and carries accepted solutions into later natural-language updates.

Watch the real workflow Explore retained results

Natural language + filesExact, scalar, and Pareto searchPersistent live statePrivate-network deployment

Captured from the deployed Harness

One clean English conversation, from initialization to a fast live update.

The recording replays a completed CNC scheduling session captured under the earlier LiveOpt name. It starts from public data, then applies one small update through a warm restart; playback makes no provider call.

Historical recording of the Harness (then named LiveOpt) receiving an English optimization problem and public data, selecting Pareto search, producing an initial result, then applying a small update with a fast warm restart. — 01 Problem + data02 Route choice03 Initial Pareto set04 Small update05 Fast warm restart

The deployed product surface

The numerical work stays visible while the agent stays quiet.

MAPLE adds purpose-built controls to the Harness instead of returning a long text-only answer.

Request and public data

Attach CSV, JSON, spreadsheet, PDF, or text sources to the same natural-language request.

Minimize weighted tardiness and energy as separate objectives.

+ Document jobs.csv · machines.csv

Choose the solving route

The agent asks when both exact and evolutionary routes are plausible. The user keeps control of the numerical method.

Exact solverEvolutionary / NSGA-II

Live numerical progress

Generation progress is polled from the local runtime. The model does not spend tokens narrating numerical search.

Evolutionary search78 / 100

100feasible18Pareto47reused1m 39selapsed

Inspectable result bundle

Each accepted turn keeps the solution set, final population, objective history, conclusion, and TSS workspace.

solutions.csviteration_history.csvresult_bundle.jsonliveopt_results.zip

Explore the retained output

Inspect initialization and the accepted live update without provider access.

Loading the published case bundle...

Accepted statet001

Natural-language request

Initial bi-objective schedule

Pareto solutions--

Final population--

Executed generations--

Restart action--

Final solution set

Pareto trade-offs

PopulationPareto set

Representative solution

Machine schedule

Inside MAPLE Harness

A compact agent loop around a fixed numerical runtime.

The model handles problem semantics and bounded updates. Typed operators, solver execution, validation, state persistence, and exports remain deterministic application services.

Prepare public data

Uploaded sources become typed public tables and bounded document context.

Build the Workbench

The model writes problem semantics; TSS binds fixed initialization, crossover, mutation, repair, and solver routines.

Solve in the background

Exact, scalar evolutionary, and Pareto search stream numerical progress without model polling.

Check before acceptance

Syntax, contract, smoke, feasibility, and output checks gate every state transition.

Adapt accepted state

LSM grounds references, while the runtime selects full restart, warm repair, or population transfer.

Inspect and export

Every turn exposes objective traces, solutions, populations, schedules, and portable result bundles.

Staged release

The product preview is public; the implementation is private during review.

The interactive walkthrough and retained case remain openly available on this page. The isolated application and research repositories will be released with the paper artifacts.

Replay the workflow Inspect the case

Public previewProvider-free

Interactive product walkthrough
Accepted solution history
Objective and Pareto visualizations
Sanitized TSS artifact view

No provider keyNo hidden evaluatorNo research cache