Skip to main content

When to use files vs internal docs

TypeBest for
File UploadsExisting content you have as a file: PDFs, internal runbooks, exported KB articles, CSVs of FAQs
Internal DocsNew content you want to write in the dashboard, AI guidance, internal policies
Both end up in the same knowledge base and are searched the same way at runtime.

File uploads

From Knowledge > File Uploads in the dashboard, click Upload and pick one or more files.

Supported file types

TypeMIME type
Plain texttext/plain
Markdowntext/markdown
CSVtext/csv
HTMLtext/html
PDFapplication/pdf
JSONapplication/json
XMLapplication/xml, text/xml

Limits

  • Maximum file size: 10 MB
  • Files are processed immediately on upload — no manual indexing needed

When to use files

Files are ideal for content that isn’t available on a public URL:
  • Internal runbooks and SOPs
  • Product specs and release notes
  • Training materials and FAQs
  • Exported knowledge base articles
  • Pricing pages, sales playbooks, support scripts
To update a file, delete the old version and re-upload. The old chunks are removed and the new content is re-ingested.

Internal Docs

From Knowledge > Internal Docs, you can write knowledge articles directly in the dashboard. Internal docs are good for:
  • Quick answers to common questions (“Yes, we offer annual billing — contact [email protected] to switch.”)
  • Internal policies or procedures
  • Instructions that don’t belong in your public help center
  • Custom guidance for your AI agent (“Always recommend the Pro plan for teams larger than 10.”)
Internal docs support plain text and Markdown. They’re chunked and embedded the same way as any other content source.

How content is processed

All content — files, internal docs, URLs, integrations — flows through the same pipeline:
  1. Chunking — split into parent chunks (~2000 chars) and child chunks (~500 chars) with overlap
  2. Embedding — converted to vector embeddings via Voyage AI for semantic search
  3. Storage — stored in your org’s knowledge base, linked to your organization
When the agent answers a question, it searches across all content sources simultaneously. Files, docs, and URLs are treated equally.

Deleting

Click the trash icon next to a file or doc to delete it. Deletions remove the embeddings immediately — the AI stops surfacing that content on the next query.

Where to go next

URL Sources

Crawl your help center or docs site automatically.

Help Center Articles

Write articles in your hosted Halo help center — they auto-ingest as knowledge.

Learning Rules

Filter what gets learned from synced integration data.