What Millwork is

Millwork gives your application one API for the models your organization may use. You save models, send tasks with your data rules and budget, and keep a receipt of each run’s outcome. A completed live run also returns the model output for your application.

​

The idea

Millwork exists so you can govern your AI workflow end to end. Think of it as an AI strategy envelope: you decide what may run, set the rules for each task, and learn from the receipt it leaves. The models, providers, and agents inside the envelope can change; the way you decide, size, run, measure, and improve stays the same. The Millwork overview walks through those five steps.

​

What you get

• One API for every model you save. Add models from your organization’s catalog, whether they come from your own provider accounts or are provided by Millwork. You send every task the same way, whichever model runs it.
• A choice governed by your request. You set data rules, a model-usage budget, and a runtime limit. Millwork picks a ready saved model allowed by your data rules. Read how the budget works.
• An output check. A basic check confirms only that output is present. It does not judge quality. You can attach a stricter check of your own.
• A record of each run. A completed live run returns the model output as its result. Every run that ends leaves a receipt with its outcome, recorded model usage, and checks. When a live run on a catalog model completes, its records also identify which model and provider answered. See what each record contains.
• A dashboard. Manage saved models, checks, runs, receipts, API keys, members, and Usage & spend for your organization.

​

Bring any verifier logic

Connect rules, evaluator models, compliance checks, proof engines, or completed workflow results through one output check endpoint. Your service makes the check; Millwork records its verdict, quality score, and any named results with the governed run. Your run policy can then decide what happens next. The result is your service’s claim. If the graded agent can write the evidence, use records or a signed grade that it cannot alter. Candidate output → your verifier → verdict, score, and named checks → run receipt. Keep your logic and credentials in your service. You can reuse an evaluator you already trust or combine several checks for one decision. Finish long-running work before Millwork calls the endpoint, then return a bounded result. Choose a recipe for your decision and follow Connect an output check for local testing, connection, and a separately approved run.

​

What you pay

Every live run, one submitted to call a model, has a platform fee. Model usage is a separate cost, and who bills it depends on where the model comes from. A test run (the API calls it Echo, mode: "echo") does not call a model, so it has no platform fee and no model cost. Read Model access and billing for the cancellation rules.

​

The two names you will meet

Millwork is the product you sign up for, pay for, and read about here. Solver is the part of Millwork that runs your task. The API and dashboard use that name. They are not two products. You do not sign up for, pay for, or configure them separately. One organization API key is all your application needs. You may also meet the name SolverAPI in the API and in names such as SOLVERAPI_API_KEY; it is the same product.

​

What Millwork adds to a model call

Along with calling a provider, Millwork records the model selection, model usage, and output checks for your application. You can inspect those records without building a separate reporting flow for each provider.