Skip to main content
Admin › Monitoring › File Logs
File Logs is a security monitoring feature for files with guardrail analysis results.
  • Admins get a unified view of uploaded file sensitivity classification, blocking status, and security check results.
File Logs main screen

File Logs main screen — filter bar, table, source/classification badges

Not all uploaded files appear here. Only files with classification results or guardrail blocks are shown in the log.

​
Document Sensitivity Classification

File guardrails are an Enterprise-only feature and are off by default. Turn on File Upload Guardrails in Admin › Settings › Guardrails, then enable the applied scopes (Chat · Knowledge · Project) and each individual check. Document Classification and NSFW Image Detection also require a model — without one, files pass without being checked.
When Document Classification is on, an LLM classifier reads the uploaded file’s text and determines its sensitivity. The table below is the built-in classification scheme. Categories can be added or removed and their action (allow / flag / block) changed under Admin › Settings › Guardrails › Document Classification, so actual handling follows your instance settings. Each classification result includes a confidence score (0–100%) and the classification reason. Files with no extractable text or a failed classification are recorded as UNCLASSIFIED, and no status badge is attached to them.

​
Guardrail Check Types

The checks you have enabled apply according to the file type.
LLM-based sensitivity classification. Analyzes document text to assign PUBLIC ~ RESTRICTED grade.
  • Target: All files with extractable text (PDF, DOCX, TXT, etc.)
  • Result: Classification grade + confidence + reason
  • Action: By default CONFIDENTIAL → flag, RESTRICTED → block

​
Viewing Logs

  • 20 entries per page
  • Total file count shown at top
  • Page navigation controls when over 20

​
Detail View

Click the arrow on the right of a table row to expand it and show the file’s guardrail analysis.
File detail — guardrail analysis result

File detail — file info, uploader, source, guardrail detail JSON

Items shown:
  • File info: Filename, content type, size (KB)
  • Uploader info: Name, email, upload time
  • Source: Upload path badge
  • Classification result:
    • Sensitivity grade + confidence (%)
    • Classification reason
    • Classification model used
    • Error message on errors
  • Guardrail Details (when a guardrail block record exists): block detail JSON
  • EXIF Stripped (when the EXIF check ran): Yes or No
  • NSFW Check (when the NSFW check ran): the model’s verdict

​
File Processing Flow

Files blocked by macro detection or NSFW image detection are rejected at upload, so they never appear in File Logs. File Logs shows files that have a classification result recorded, plus files that were blocked by the text content guardrail and left a block record.

​
Use Cases

  1. Set the Category filter to CONFIDENTIAL only
  2. Review the flagged file list
  3. Confirm classification reason and confidence in the expanded detail
  4. Adjust policy in Guardrail Settings as needed
  1. Use the Source filter to view upload status by Chat / Knowledge / Project
  2. If a source has many blocks, inspect that workflow
  3. Review Knowledge upload classification results to manage RAG data quality
  1. Set the Status filter to Blocked. Files blocked by document classification carry no status badge, so check them separately with the Category filter set to RESTRICTED
  2. Confirm uploader and reason for blocked files
  3. Review guardrail detail JSON in the expanded detail
  4. Consider revising guardrail rules if false positives

​
Best Practices

  • Periodic monitoring: Review file logs at least weekly to detect anomaly patterns
  • Guardrail integration: When blocks/flags occur frequently, review Guardrail Settings
  • False positive management: When CONFIDENTIAL/RESTRICTED classification confidence is low, consider changing the classification model
  • Source analysis: Watch for patterns where sensitive files concentrate in a specific source