Cut your AI bill. Control the rest.

You use Darcy Auto, which picks Frontier or Swift for each request, or you pin Frontier or Swift directly. Behind each tier, Innovative runs one or more vetted third-party models and changes them as better or cheaper options appear. We don’t publish the roster, but every provider meets our Provider Standards.

Automatically, with nothing for you to choose. For each tier, we designate one comparable model from each of Anthropic and OpenAI, chosen by capability rather than by name. For every request, we calculate what it would have cost at each model’s standard published prices and use the higher of the two as your baseline. Input, cached input, and output tokens are each priced at the model’s standard rate for that type, and promotional and batch prices are never used. Every request is priced and recorded the moment it’s made, so later price changes never rewrite your history. Because routing can send requests to different models, you may see fewer cache hits than with a single provider, and savings are always calculated on the caching you actually got. Darcy Dispatch shows the current reference models, their prices, and exactly how every savings figure is calculated.

Auto judges every request on its own. Hard work, such as a failed tool call, an error to diagnose, or a plan to make, goes to Frontier. Settled, mechanical steps stay on Swift. Behind each tier, our team picks the best price for performance from commercial models, open-weight models, and spare GPU capacity, all held to our Provider Standards.

They shouldn’t. A new model joins a tier only if it matches what that tier serves today, at a better price, after it’s scored on independent intelligence, coding, and agentic benchmarks and tested on real agent workloads. Auto escalates the moment a task gets difficult, you can pin any workload to Frontier, and admins can turn tiers on or off for each team.

Our team scores it on independent benchmarks, tests it on real agent workloads, and verifies its terms. It joins a tier only if it matches what that tier serves today, at a better price, and we send you a short review of what it means for your organization. If it’s a new Anthropic or OpenAI model, it can replace that lab’s reference model in your savings benchmark right away.

Then there’s no fee on that request. You pay only for the inference at Darcy Dispatch published rates.

Yes. Darcy Dispatch tags every request automatically by team, project, or client. You can break spend and savings down by person, team, tier, tag, or client over any window and export it to CSV for your ERP or billing system, for internal chargeback or client invoicing.

Automatically. Dispatch inspects each request as it arrives and tags it from its key, its guardrail, and the request itself, including requests from tools like Cursor and Copilot. Teams inherit tags from their guardrail, and when a shared tool or agent works across clients or projects, Dispatch classifies each request to decide which one it belongs to. Nobody tags anything by hand, and your own apps can add tags to individual requests as well.

Yes. Darcy Dispatch lets you set monthly budgets by person or team. The owner is notified as spend approaches the cap, and requests stop when it is reached. Dispatch also shows run rate, projected month-end spend, and credit runway. Your TAM reviews the forecast with you every month.

Yes. Auto picks the tier for each request by default, and you can pin a workload to Frontier or Swift when it needs one. Admins can also turn tiers on or off for each team. Pinning can reduce savings, and your TAM will walk you through the tradeoff.

A guardrail is a policy assigned to a key or team. It catches personal data and secrets before any model sees them and redacts or blocks them, and it carries the tags that team’s requests inherit. Every catch is recorded in the audit log.

No. Innovative doesn’t train on your content, and we only route to Zero Data Retention endpoints at providers that are barred from training on it. Providers store none of your prompts or completions. If you bring your own provider keys, your agreements with those providers govern instead.

Yes. Turn on U.S.-only processing in Darcy Dispatch, and every request is processed and stored in U.S. data centers. You can turn it on or off anytime. Because U.S.-only routing narrows our options, you’ll save less while it’s on. Darcy Dispatch itself, and all your records, always stay in the U.S.

No. Commitments are optional, and you can start without one. A 12-month commitment with a minimum monthly credits purchase earns better pricing, so you keep more of the savings Dispatch delivers. Your TAM can model the options against your usage before you decide. Without a commitment, you can stop using the service whenever you like, with no notice. Unused Credits simply expire.

Yes, with or without a commitment, for some or all of your requests. Add your keys in Darcy Dispatch. Requests through your keys use no credits, you pay your providers directly, and we charge only our share of the savings we calculate, monthly. Your agreements with those providers govern how they handle your data. You can switch between your own keys and our inference anytime.

Service pauses until you add credits. Automatic recharge prevents that for production traffic, and you’re never invoiced for usage beyond your balance.

Yes, at any time, because model prices move constantly. Darcy Dispatch always shows the current rates and the current baseline, and calculates your savings on every request against today’s prices, so you never have to wait for a notice to know where you stand.

U.S.-based businesses whose users are 18 or older and located in the United States. MRS is not available for use from, or for personal data of people in, the EU, the UK, Switzerland, Australia, or sanctioned countries. MRS is governed by the Innovative MRS Terms, our Standard Terms, and the Darcy Dispatch Account Terms, Data Processing Addendum, and Acceptable Use Policy.