Route every step to a model that fits, under your provider contracts. } subtitle="LLM routing for delegated work. Your application hands over a Task and gets one result back. Duale AI runs each part of the Task on a model you enabled." primaryCta={{ label: "Plan your first use case with us", dialog: "contact" }} secondaryCta={{ label: "Read the model routing guide", href: "https://duale.ai/en/docs/model-routing.md" }} />

Connect the model providers your company already uses, and the models you host yourself. [See what each provider supports](https://duale.ai/en/docs/model-routing/provider-capabilities.md). Your team decides which models join your model pool. Duale AI never calls a model outside it. Your platform team stores each provider key once, encrypted in the Platform. Duale AI calls the models on behalf of your applications.

Reading a long document can go to a model built for that, and a simple update to a fast, economical one. Legal analysis, finance, code and multi-step work with Tools can each run on a different model you enabled. [See which skills you can ask for](https://duale.ai/en/docs/model-routing/task-capabilities.md). ```mermaid flowchart LR accTitle: Work and billing for one delegated Task accDescr { Your application delegates one Task to Duale AI. Duale AI runs each part of the work on a model you enabled at your own providers, then returns one result to your application. Your providers bill inference to your finance team under your contracts. Duale AI bills a fixed price per Task. } App["Your application"] Duale["Duale AI"] Models["Models you enabled, at your providers"] Finance["Your finance team"] App -->|delegates one Task| Duale Duale -->|each part of the work| Models Duale -->|one result| App Models -.->|inference, under your contracts| Finance Duale -.->|fixed price per Task| Finance ``` Your finance team sees inference on your provider invoices and a fixed price per Task from Duale AI, whichever models did the work.

Usage analytics records which model ran each Task. [LLM Coach](https://duale.ai/en/product/usage-analytics.md) compares the models in your pool, so you decide each change with evidence in hand. **Illustrative example.** Your catalogue team delegates product-sheet translation Tasks every day. - **Day 1.** A new model is released. Your platform team adds it to the model pool for evaluation and runs last week's translation Tasks on it. - **Day 2.** LLM Coach shows the new model beside your current ones. The team enables it. - **Day 4.** Your reviewers correct more product names than usual. The team switches back to the previous model. - **All week.** The catalogue application sent the same Task and received the same result format. Nobody changed its code.

The retry can run on the same model or on another one. With models from more than one provider enabled, the work can continue at one provider while another is down.

You get one due diligence memo with every finding sourced. Each data room document goes to a model that fits it. You get one evidence map with every claim sourced. Reading long papers and writing the synthesis can each run on the model that fits them. Your lawyers get one list to act on, each point quoted from its source. Each review runs only on models your company enabled.

A standard Task costs €0.10 and a fast Task €0.20, excluding VAT. [See pricing](https://duale.ai/en/product/pricing.md) Your model providers bill you directly for inference. Use the managed service, hosted in the European Union, or run the Platform in your own infrastructure.

## Model routing questions ### Can one job stay on a single approved model Yes, and the rest of your work can still use every model you enabled. [Check in Usage analytics which model ran each Task](https://duale.ai/en/product/usage-analytics.md), and [see how to isolate work bound to one model](https://duale.ai/en/docs/model-routing/architecture-patterns.md#isolate-a-strict-workload). ### Are our model providers Duale AI subprocessors No. They process your data under your own agreements, not under Duale AI's. [See how the Trust Center covers the providers you choose](https://duale.ai/en/product/security.md), and [read the AI services section of the subprocessor list](https://duale.ai/en/legal/subprocessors.md#artificial-intelligence-services). ### Can anyone read our provider keys once they are stored No. A stored key is never shown again, even to the people who manage your models; they can only replace it. [Read how a stored key stays hidden](https://duale.ai/en/docs/model-routing/provider-capabilities.md#access-needed-to-change-a-route). ### Can our models read the images in our documents Yes. Charts, photos and scanned pages in your documents reach a model you enabled that reads images, so the result draws on them too. [See how agents read your documents with Libraries](https://duale.ai/en/product/libraries.md), and [see what decides whether an image reaches a model](https://duale.ai/en/docs/model-routing/images.md#what-decides-whether-an-image-reaches-the-model). ## Related content - [Model routing: route selection and configuration](https://duale.ai/en/docs/model-routing.md) - [Provider capabilities on a Model Route](https://duale.ai/en/docs/model-routing/provider-capabilities.md) - [Model routing contract and limits](https://duale.ai/en/docs/model-routing/contract-and-limits.md) - [Product: Task execution and platform capabilities](https://duale.ai/en/product.md) - [Procurement: every supplier, invoice and bid checked](https://duale.ai/en/use-cases/procurement.md) - [AI that takes on real work, with and for your people](https://duale.ai/index.md) --- ## Sitemap See the full [Markdown sitemap](https://duale.ai/sitemap.md) for all pages.