LintPDF LintPDF

AI troubleshooting

FAQ, common errors, and the full error-code reference for AI inspections.

AI troubleshooting

Everything that can go wrong with AI inspections, plus answers to the questions we get most often.

FAQ

General

What are AI credits?

AI credits are the billing unit for AI-powered inspections. Text-tier inspections (text analysis, barcode decode) cost 1 credit. Vision-tier inspections (vision models, regulatory compliance) cost 2 credits. Credits are purchased separately from your plan subscription.

How do I enable AI features?

AI features are in invite-only alpha. Email sales@lintpdf.com with your account ID and use case. Once enabled, configure AI categories under Settings → AI Inspections in the dashboard.

Do AI features replace the core engine?

No. The core engine runs 500+ deterministic, rule-based checks plus the 91-check PDF/X-4 conformance suite (ISO 15930-7), and is not affected by AI features. AI inspections run alongside core engine checks and produce additional findings in the same report — the source field distinguishes them ("engine" vs. "ai").

Which plans support AI features?

All paid plans: Starter, Growth, Scale, and Enterprise. Credits are purchased separately from your subscription.

Credits & billing

How much do AI inspections cost?

Text-tier inspections cost 1 credit ($0.12 pay-per-use, or less with volume packages starting at $0.10/credit). Vision-tier inspections cost 2 credits. See the pricing page for package options.

Do credits expire?

No. Credits purchased through packages never expire. Pay-per-use credits are billed as consumed.

What happens when I run out of credits?

AI inspections are skipped. Core engine checks continue normally. The report includes an info note saying which AI inspections were skipped. Configure the ai.credits.depleted webhook to be notified.

Can I set a spending limit?

Yes. For pay-per-use billing, set a monthly cap under Settings → AI Billing. When the cap is reached, AI inspections are skipped for the remainder of the billing cycle.

Checks

Why did my AI check fail?

Common reasons:

  • Insufficient credits — balance was zero. Purchase credits or enable auto top-up.
  • Circuit breaker tripped — vision capacity was constrained; retry in a few minutes.
  • Category not enabled — enable under Settings → AI Inspections.
  • File too large — AI has a 100 MB limit. Core engine checks still run.
  • Unsupported format — some AI inspections require rasterised page content; encrypted or malformed PDFs may fail here.

What is the circuit breaker?

Vision-tier inspections use a circuit-breaker pattern. When vision infrastructure is at capacity the breaker trips and vision inspections are gracefully skipped. Text inspections and core engine checks are unaffected. The breaker auto-resets once capacity recovers.

How accurate are AI findings?

Every AI finding includes a confidence score (0.0–1.0). Findings are only reported above a configurable minimum threshold (default 0.75). Regulatory compliance inspections typically have confidence > 0.90. Adjust the threshold under Settings → AI Inspections.

Can I adjust the confidence threshold?

Yes — Settings → AI Inspections. Lower thresholds produce more findings but may include false positives. Higher thresholds reduce noise but may miss edge cases.

Configuration

How do I set up brand colors?

Settings → AI Brand → Color Palette. Add hex values + set the Delta E tolerance (default 5). See Brand config.

How do I upload reference logos?

Settings → AI Brand → Logos. Upload PNG, SVG, PDF, or EPS variations. More variations = better matching.

Can I add words to the spell-check dictionary?

Yes — Settings → AI Brand → Dictionary. Brand names, product names, and technical terms get exempted from the spell checker.

How do I use AI presets in a ruleset?

Include ai_preset in your submit request or configure AI categories in a custom ruleset. See Presets.

Technical

Text vs. vision tier?

Text-tier runs text analysis (spell check, barcode decode, palette matching) with sub-second latency at 1 credit each. Vision-tier runs vision-based analysis (regulatory panels, logo matching, image quality) with 1–5 s latency at 2 credits each.

Does AI processing affect submit latency?

AI inspections run in parallel with core engine checks. Text-tier adds minimal latency; vision-tier may add 1–5 s. The core engine portion of your report is available as soon as core processing completes, even if AI inspections are still running.

Are AI findings in PDF reports?

Yes. AI findings appear in PDF, JSON, and HTML reports alongside core engine findings. In PDF reports they carry an “AI” badge.

Can I use AI via API only?

Yes. All AI features are available via the API. The dashboard is just a UI for configuration.


Error code reference

AI-specific error codes that may appear in API responses or report info notes.

Error codeWhat it meansResolution
ai.credits.insufficientCredit balance is zero or insufficient for the requested inspectionsPurchase credits or enable auto top-up
ai.credits.depletedCredits ran out mid-processing — some inspections were skippedTop up credits; skipped inspections noted in the report
ai.circuit_breaker.openVision capacity constrained — vision inspections temporarily unavailableRetry in a few minutes; text inspections unaffected
ai.category.disabledRequested AI category is not enabled on this accountEnable under Settings → AI Inspections
ai.not_enabledAI features are not enabled on this accountRequest access via sales@lintpdf.com
ai.preset.not_foundThe specified preset ID does not existGET /api/v1/ai/presets to list valid presets
ai.file.too_largeFile exceeds the 100 MB AI limitReduce file size; core engine checks still run
ai.rasterization.failedPage could not be rasterised for a vision-based inspectionCheck for encryption, malformed structure, or unsupported features
ai.brand.no_paletteBrand palette inspection requested but no palette configuredConfigure under Settings → AI Brand
ai.brand.no_logosLogo matching requested but no reference logos uploadedUpload under Settings → AI Brand
ai.model.timeoutAI model processing exceeded timeout thresholdRetry; if persistent, contact support