Tutorial: Configure AI capture
Connect a provider, pick an AI service, and analyse your first document with AI.
In this tutorial you'll set up AI capture from scratch and run it on a real document. You'll choose where the AI runs — fully on-device, or through your own cloud account — pick the AI service that matches your document type, click Analyse with AI, and review the values it pulled out. By the end you'll have one document with AI-extracted fields that you've checked and corrected, and an AI engine that's ready for the rest of your batch.
Allow about ten minutes. You'll need one representative document (an invoice, a receipt, a passport — whatever you actually process) and, if you want to use a cloud engine, an API key from your own Anthropic (Claude) or OpenAI (GPT) account.
Bring your own AI — Scanix never sees your key or your documents
Cloud calls go straight from your machine to your own provider account. Scanix is never the account holder, never receives your documents, and never bills you for tokens. You can also run entirely on-device, with no cloud egress at all. The choice is yours, and you make it once. See AI providers, routing & privacy.
What you'll have at the end
- An AI engine configured the way you want it — on-device, your own cloud key, or both.
- One document run through Analyse with AI, with its fields filled in automatically.
- Those fields reviewed and, where needed, corrected — a clean, AI-extracted document.
- Corrections that Scanix keeps, so the next document of the same type starts from a better place.
Step 1 — Choose where the AI runs
AI capture can run two ways, and you pick whichever fits your privacy and connectivity needs. You wire up the engines on the AI Providers tab and choose your app-wide default on the AI Services tab — both under Settings.
On-device means the AI runs on the same machine as Scanix, using installed local models. Nothing leaves your computer — there's no cloud egress and no API key to manage. It's the right choice for air-gapped sites or sensitive documents.
Open AI Services
From the left navigation, open Settings, then the AI Services tab. The app-wide Default provider & routing lives here.
Make the local engine your default
Under Default provider & routing, set Default provider to On-device (local) (its note reads "No cloud egress; uses installed local models"). Set Default routing policy to Local first, which keeps everything on-device unless you later add a cloud key.
Screenshot
The AI Services tab with Default provider set to On-device (local) and Default routing policy on Local first. — shot tutorials-configure-ai-capture-01
Install a local model first
On-device extraction uses installed local models. If none are installed yet, install one from Local Models on the AI Providers tab before you run the analysis. With Local first routing, Scanix tries on-device and only escalates to a cloud provider if you've configured one.
A cloud engine sends the page to your own Anthropic (Claude) or OpenAI (GPT) account for the highest-accuracy extraction. You bring the key; Scanix never sees it.
Open AI Providers
From the left navigation, open Settings, then the AI Providers tab.
Add your API key
Under Cloud — bring your own API key, find your provider's row and paste your key into its API key field, then click Save. The status pill changes to Configured (showing the last four digits). You can use Test to send a quick connectivity check.
Set it as your default
Switch to the AI Services tab. Under Default provider & routing, set Default provider to Anthropic (Claude) or OpenAI (GPT). To force every run through that cloud provider, set Default routing policy to Cloud only; to keep work on-device first and only escalate to the cloud when needed, set it to Local first.
Screenshot
On the AI Providers tab, the Cloud — bring your own API key section with a provider row expanded, the API key field filled, and the configured status shown. — shot tutorials-configure-ai-capture-02
That's the only setup that's global. From here on, the document type drives everything.
Step 2 — Pick an AI service
An AI service (also called an AI module) is a pre-built, typed schema for one class of document — Invoices AI, Receipts AI, Identity Documents AI, and more. Picking the service tells Scanix exactly which fields to look for and how to validate them.
Open a document and find the AI picker
Bring a document into the Viewer (scan it or import it). In the right-hand panel, you'll see a picker for choosing an AI service for this document.
If a template already chose the service
If you opened the document under a template that pins an AI service, the choice is already made for you and the picker is shown read-only — skip ahead to the next step. To have AI build a template for you from a single sample, see Smart Templates.
Choose the service that matches your document
Select the service for your document type — for example Invoices AI for a supplier invoice, or Receipts AI for an expense receipt. Each service lists the fields it extracts, so you can confirm you've picked the right one before running it.
If your document doesn't fit any of the pre-built schemas, choose Natural Language Query instead and describe the fields you want in plain English — for example, "Find the customer email, the order number, and the shipping address."
Screenshot
The Viewer's right panel showing the AI service picker with Invoices AI selected, and the service's field list beneath it. — shot tutorials-configure-ai-capture-03
Step 3 — Analyse with AI
With a service selected, the primary action button in the Viewer's right panel becomes Analyse with AI. One click runs OCR and the AI service together in a single pass.
Run it
Click Analyse with AI. Scanix reads the page, then runs your chosen service against the recognized text and fills in the fields. The button shows its working state while the extraction runs.
Screenshot
The Viewer mid-run: the primary button reads Analyse with AI and shows its busy state while the fields populate. — shot tutorials-configure-ai-capture-04
Watch the fields fill in
When the run finishes, the Fields panel on the right shows each field the service extracted, with the value it found. Dates are normalised, amounts are structured, and where a service can cross-check itself — for example, that line items sum to a total — it does.
Where the document was sent
With an on-device engine, the page never left your machine. With a cloud engine, it went straight from your computer to your own provider account. Either way, Scanix didn't store your document on a server or charge you for the run. The full picture is in AI providers, routing & privacy.
Step 4 — Review and correct the fields
The AI does the heavy lifting; you stay in control. Read through the extracted values and fix anything that's off.
Check each field against the page
In the Fields panel, scan the values. Click a field to see where it came from on the page. Most will be right; correct any that aren't by editing the value inline.
Re-run if you've changed the source
If you edited the recognized text or switched the service or routing, click Analyse with AI again to extract from the updated page. Your manual field edits are preserved across a re-run — only the AI-filled values are replaced.
That's your finished result: one document, read and extracted by AI, with every field reviewed and corrected. It's ready to process and export like any other Scanix document.
How your corrections improve future runs
Your corrections aren't just for this document. When you fix a field, Scanix keeps that confirmed value — and a manual correction always survives a later re-extraction, so it's never silently overwritten. The next time you analyse a document of the same type, Scanix can draw on a few of your most recent confirmed examples to guide the AI toward the answers you've already shown it. In practice, the more you correct, the less you have to.
You're set up
You've configured an AI engine, picked a service, run Analyse with AI, and reviewed the result — and Scanix is now keeping your corrections to make the next run better. Point it at the rest of your batch, or wire AI into a whole pipeline with Job Groups.
Next steps
AI Services catalogue
The document services Scanix can read, and the fields each one extracts.
Analyse a document with AI
A focused reference for the Analyse with AI button and the Fields panel.
AI providers, routing & privacy
On-device vs. cloud, routing policies, and exactly what leaves your machine.