Skip to content

Documents and Files

Uploading documents is the fastest way to equip your AI agent with deep domain knowledge regarding your products, operating procedures, contracts, price lists, and technical guides.


Supported File Formats and Limits

FormatExtensionMaximum SizeRecommended Use Case
PDF Documents.pdfUp to 50 MB per fileCatalogs, user manuals, spec sheets, policy documents
Microsoft Word.docx, .docUp to 30 MB per fileStandard Operating Procedures (SOPs), employee handbooks, terms
Plain Text.txt, .mdUp to 10 MB per fileSupport notes, troubleshooting logs, FAQs
Tabular Data.csvUp to 20 MB per fileProduct price lists, inventory matrices

How to Upload Files

  1. Navigate to Knowledge Base → Files
    From your agent’s dashboard sidebar, click the Files tab.

  2. Drag & Drop Your Documents
    Drag one or multiple files directly onto the upload zone or click Browse Files.

  3. Automated Parsing & Chunking
    Once the upload completes, our ingestion pipeline initiates:

    • Text extraction and removal of unneeded visual artifacts.
    • Semantic chunking that preserves table formatting and heading hierarchy.
    • High-dimensional vectorization into your isolated tenant index.
  4. Verify Training Status
    When the document badge switches to Trained (green), the agent can immediately query the newly ingested knowledge.


Best Practices for Document Optimization

To ensure the highest retrieval accuracy:

  1. Structured Headings: Documents with clean heading hierarchies (H1, H2, H3) allow our semantic parser to accurately link topics to context.
  2. Readable Column Headers: For tables featuring specifications or pricing, ensure columns have clear names (e.g., Product SKU, Wattage, Price Incl. VAT).
  3. Deprecate Outdated Versions: When issuing a new revision of a price list or manual, delete or replace the previous file to prevent conflicting responses.

Storage & Multi-Tenant Security

All raw files are encrypted at rest (AES-256) and stored in Cloudflare R2 object storage with strict tenant-isolated access controls:

  • No other workspace can access or query your uploaded files.
  • Document text is never forwarded to public foundation models for training.
  • Deleting a file instantly purges both the raw binary and all corresponding vector embeddings permanently.