AI services & automation
Activate document services, set the default provider and routing, the confidence threshold, and privacy.
Two Settings tabs control everything AI in Scanix Desktop. AI Providers is where you wire up the engines — install on-device models, connect your own cloud keys, pick the Smart Template engine, and set the reasoning policy. AI Services is the control room for the document-capture catalogue — which services are active, the default provider and routing, the auto-accept confidence bar, and the privacy posture. This page is a reference for every option on both tabs.
Open them from Settings in the left navigation, then choose AI Providers or AI Services in the tab list. On the AI Services tab, every control saves the instant you change it — there is no Apply step. On the AI Providers tab the simple toggles save instantly too, but the cloud API key rows and the Smart Template — default AI engine picker have their own Save button you click to store the change.
Bring your own AI
Cloud calls go straight from your machine to your own provider account (Anthropic or OpenAI). Scanix is never the account holder and never receives, transmits, or stores your key. You can also run entirely on-device with no cloud egress at all. See AI providers, routing & privacy.
AI Providers tab
This tab is organised top to bottom into Pipeline, Local Models — runs on this machine, Cloud — bring your own API key, Smart Template — default AI engine, and Reasoning policy.
Screenshot
The AI Providers tab: the Pipeline toggle, the Local Models — runs on this machine list with Install buttons, the Cloud — bring your own API key section, and the Smart Template — default AI engine card. — shot settings-ai-01
Pipeline
| Control | Type | What it does |
|---|---|---|
| Auto-process documents after OCR | Toggle | When OCR completes, automatically run AI classification (route to template) and field extraction. Per-template AI flags still apply — templates without AI processing enabled are unaffected. |
Local Models — runs on this machine
On-device models are pulled into the app's local model folder and run natively — no external runtime required. Local models satisfy HIPAA / PHI workflows automatically, because document data never leaves the device.
| Control | What it does |
|---|---|
| Install | Download a model from the catalogue. The button shows download progress, and you can cancel a download in progress. |
| Installed | Marks a model that is already on disk. A Delete button removes it. |
| Loaded in memory banner | Shows which local model the cache currently holds, with an Unload control. |
| Remove all installed models | Deletes every installed local model to reclaim disk. Gated behind a typed confirmation so a misclick can't wipe your model collection. |
Reasoning-capable models carry a badge in the list. Incomplete installs (downloads killed mid-stream) are surfaced in their own panel so you can recover them.
Cloud — bring your own API key
Each cloud provider has its own API key row. Calls run directly from your machine to the provider under your own account and contract.
| Control | What it does |
|---|---|
| API key field | Paste your provider key. A show/hide control reveals or masks it. |
| Save / Replace | Stores the key (the button reads Replace once a key is already set). |
| Test | Sends a tiny ping to the provider to verify connectivity. Available once a key is saved. |
| Clear | Removes the saved key. |
| Status pill | Configured (with the last four digits) or Not configured. |
Key storage is Windows-only
Saved keys are sealed at rest with Windows DPAPI. On a configured machine you can also point the key at an environment variable (ANTHROPIC_API_KEY or OPENAI_API_KEY) to override the saved field; a key sourced this way shows a from env marker and is cleared in your OS environment, not in this panel.
Cloud AI is a licensed feature
Bring-your-own cloud keys are gated by your plan. If your tier doesn't include it, the section still appears but is dimmed with an upgrade prompt above it.
Smart Template — default AI engine
Picks the engine the Smart Template wizard uses when you start a new analysis, so you skip the engine-selection step.
| Control | What it does |
|---|---|
| Provider | The AI provider. Each option is tagged local or cloud, and cloud providers without a key show (key needed). |
| Model | The vision-capable model for the chosen provider. |
Reasoning policy
Controls how aggressively reasoning-capable models are used. Per-template overrides live on each template's own AI config.
| Option | Behaviour |
|---|---|
| Off | Never use reasoning, even when models support it. |
| Auto (recommended) | The model decides per call. |
| Force on with budget | Always reason — best for cross-field validation. Reveals a Reasoning budget (tokens) field. |
AI Services tab
This tab activates the document-capture services you use and tunes the automation and privacy posture. Active services appear in the Job Group and Template designers and drive AI document splitting and extraction.
Screenshot
The AI Services tab: the Document services catalogue with Enable all / Disable all and the search box, the Default provider & routing card, the Auto-accept slider, and the Privacy & fallback toggles. — shot settings-ai-02
Document services
Each service is a document-type capture module you switch on or off. The card header shows how many are active out of the 27 built-in services (for example, "26 of 27 active"), and every service lists its field count.
| Control | What it does |
|---|---|
| Enable all | Activates every service. |
| Disable all | Deactivates every service. |
| Search services… | Filters the list by name. |
| Per-service toggle | Activates or deactivates that one service. A deactivated service is refused at extraction time. |
Services are grouped into these categories:
| Category |
|---|
| Finance & Accounting |
| Procurement & Logistics |
| HR & People |
| Legal & Governance |
| Identity & Compliance |
| Other |
Default provider & routing
Applied when a Job Group or template doesn't pin its own provider.
| Option | Behaviour |
|---|---|
| Automatic | Pick the first available provider (cloud first, then local). |
| Anthropic (Claude) | Requires a configured key in AI Providers. |
| OpenAI (GPT) | Requires a configured key in AI Providers. |
| On-device (local) | No cloud egress; uses installed local models. |
| Option | Behaviour |
|---|---|
| Local first | Try on-device, escalate to cloud only if it fails verification. |
| Local only | Never send documents to the cloud. |
| Cloud only | Always use the configured cloud provider. |
| Pinned | Use exactly the pinned provider and model. |
Automation
| Control | Range | What it does |
|---|---|---|
| Auto-accept above N% confidence | Slider, 50–99% (default 97%) | Documents whose extracted fields all clear this bar export automatically; anything below is held for review in the Verification Queue. |
Privacy & fallback
| Toggle | What it does |
|---|---|
| Local-only mode (no cloud egress) | Hard-blocks every AI service from sending document content to a cloud provider, regardless of per-group routing. An enterprise data-residency hard stop. |
| Fall back to plain OCR when no service matches | When a document doesn't match any active AI service, still export its OCR text and zones instead of dropping it. On by default. |
| Locate fields on the page (beta) | When the AI reads a page image, also ask it for a short printed-text snippet near each value, matched against the OCR to place the field highlight and magnifier. Helps locate handwritten or mis-OCR'd values. |
Locate fields on the page is beta
This option is beta and off by default. It requires a cloud vision provider, so it has no effect under Local-only mode.
Next steps
AI providers, routing & privacy
How routing, cloud egress, and on-device inference fit together end to end.
AI Services catalogue
The full list of document-capture services and the fields each one extracts.
Settings overview
A tour of every Settings tab and how saving works.