Skip to main content

​
Beyond metadata

Traditional file validation checks metadata: the extension, the MIME type, the file size. These are labels chosen by the sender. They tell you what the file claims to be — not what it actually is. Content intelligence is Uplint’s ability to look beyond metadata and understand the substance of what’s entering your system.

​
Three levels of intelligence

​
Structural truth

Is this file what it claims to be? Uplint inspects the internal binary structure to verify the format is genuine. A file that says it’s a PDF but has an executable header isn’t confused — it’s deceptive. Corrupt headers, polyglot files, format spoofing — structural intelligence catches what no label check can. What it catches:
  • Executables renamed as PDFs
  • Polyglot files valid in multiple formats
  • MIME type spoofing
  • Corrupt or truncated files

​
Substantive reality

Is there meaningful content inside? A PDF with zero readable words isn’t a document. A spreadsheet with headers but zero data rows isn’t a report. An image that’s a single solid color isn’t a photo. These files pass every traditional check — correct extension, valid MIME type, reasonable size — but they’re worthless. What it catches:
  • Blank PDFs (zero readable words)
  • Empty spreadsheets (headers only, no data)
  • Single-color images
  • Whitespace-only text files

​
Semantic context (coming soon)

Does this data belong where it was sent? This is the most powerful level of content intelligence. When your infrastructure understands content semantically, it can verify that a document uploaded to insurance_claims is actually a claim form, a prescription, or a medical bill — not a random receipt, a personal photo, or someone’s homework. Modern vision models and multimodal AI make this feasible at API speed. Your infrastructure can understand what data is, the way a human reviewer would — but on every submission, instantly.

​
Why this matters

In regulated industries, content-blind systems create real risk:
  • A blank PDF accepted as a patient record creates a gap in the audit trail
  • A renamed executable in a document upload is a security incident waiting to happen
  • A vacation selfie accepted as a medical claim means your system can’t tell the difference between valid and invalid submissions
Every downstream system inherits the trust decisions made at the boundary. If your boundary is content-blind, your entire stack is content-blind.

​
How to enable it

Content intelligence is configured per file context: 
{
  "context_key": "patient_reports",
  "reject_blank_files": true,
  "reject_corrupt_files": true,
  "scan_for_viruses": true
}
  • reject_blank_files: true — Enables substantive content analysis
  • reject_corrupt_files: true — Enables structural validation
  • scan_for_viruses: true — Enables threat detection
All three are enabled by default when creating a new context.