Skip to main content

Documents Settings

Organization defaults, the category catalog, storage, and the deployment settings an administrator controls.

Overview​

Open Settings β†’ Documents. The screen needs the DOCS_MANAGE permission and is organized as four cards:

  1. AI knowledge status β€” what this deployment can actually do
  2. Organization defaults β€” what new documents inherit
  3. Storage β€” how much of the quota is in use
  4. Categories β€” the category catalog

AI Knowledge Status​

A read-only report of the deployment's capabilities, so you know why AI features behave the way they do before changing anything else.

LineMeaning
Vector search is availableMeaning-based search is working
Vector search is unavailableAnswers fall back to keyword search
An embedding provider is configuredDocuments can be prepared for AI answers
No embedding provider is configuredDocuments cannot be indexed β€” AI answers will not work
Embedding modelThe model currently in use

If either line shows a warning, see Environment Settings below.

Organization Defaults​

Three settings that apply to newly created documents in this organization.

SettingEffect
Add uploads to AI knowledge by defaultPre-selects the AI knowledge option on the upload dialog
Classify with AIDefault for the upload dialog's own Classify with AI toggle, which suggests categories, tags, and a summary. Uploaders can override it per batch
VisibilityWhether new documents start as Organization or Private
tip

Setting Visibility to Private by default suits organizations handling sensitive material β€” documents then start closed and are opened deliberately, rather than the other way round.

Storage​

Shows how much storage the organization's documents occupy, with a progress bar and a warning as you approach the limit.

Two things people are often surprised by:

  • Archived files still count. As the hint says, the usage figure "counts every uploaded file, including archived ones β€” their bytes still exist."
  • When the limit is reached, new uploads are rejected. They are not silently truncated.

If the organization has no quota, the card simply reports total usage instead.

The Category Catalog​

Categories are the curated taxonomy used to classify documents (and what AI classification chooses from). Each organization starts with eleven:

Invoice Β· Contract Β· Report Β· Policy Β· Customer List Β· Expense Β· HR Β· Legal Β· Meeting Notes Β· Specification Β· Other

The table lists each category with its color, description, and how many documents use it. You can:

ActionNotes
New categoryAdd one of your own with a name, color, and description
EditRename, recolor, or describe any category, including the built-in ones
Merge into…Move every document into another category and remove this one
DeleteOnly available for categories you created

Built-in categories are marked System. They can be renamed and merged but not deleted, so classification never breaks because a category vanished underneath it.

note

Keep the catalog small and meaningful. AI classification picks from this list, and a catalog of eighty near-identical categories produces worse suggestions than one of a dozen clear ones.

Environment Settings for Administrators​

The Documents hub works with no configuration at all: files upload, their text is read out, and they become searchable. Everything below is optional tuning applied when the platform starts, and all of it ships commented out.

warning

AI features are on by default, and it is safe to leave them on with no provider configured β€” the AI stages are additionally gated on a provider having credentials, so nothing is attempted until one exists. Set GAUZY_DOCS_AI_ENABLED=false to keep them off even where a provider IS configured. Documents reuses the platform's AI chat providers, including per-tenant keys.

AI and Knowledge​

VariableDefaultControls
GAUZY_DOCS_AI_ENABLEDtrueMaster switch for classification, summaries, and embeddings
GAUZY_DOCS_OCR_ENABLEDtrueReads scanned PDFs and images β€” one model call per page
GAUZY_DOCS_OCR_MAX_PAGES20Page ceiling per document, so one large scan cannot run away
GAUZY_DOCS_CLASSIFY_MODELAI chat default modelModel used to classify documents
GAUZY_DOCS_EMBEDDING_MODELtext-embedding-3-smallModel used to prepare documents for AI answers
GAUZY_DOCS_EMBEDDING_DIMS1536Vector dimensions β€” must match the storage column
GAUZY_DOCS_EMBED_BATCH_SIZE64How many passages are embedded per request
GAUZY_DOCS_CLASSIFY_SAMPLE_CHARS4000How much of a document is sampled for classification
GAUZY_DOCS_VECTOR_STOREbest availablePins the search backend; degrades to keyword search when unavailable
GAUZY_DOCS_RETRIEVAL_TOPK_MAX12Maximum passages returned for one question
GAUZY_DOCS_CHUNK_TOKENS512Size of each passage
GAUZY_DOCS_CHUNK_OVERLAP_TOKENS64Overlap between neighbouring passages
GAUZY_DOCS_AUTO_REINDEX_ON_MODEL_CHANGEfalseRe-prepare every document when the embedding model changes
GAUZY_DOCS_RETRIEVAL_LOG_ENABLEDtrueLogs query length, result counts, and latency β€” never the query text

Meaning-based search needs PostgreSQL with vector support. On other databases, or when that support is missing, retrieval falls back to keyword search automatically and the settings screen reports it.

warning

Leave GAUZY_DOCS_AUTO_REINDEX_ON_MODEL_CHANGE off unless you mean it. Changing the embedding model re-prepares every document in the deployment, which costs real money at your AI provider.

Uploads and Processing​

VariableDefaultControls
GAUZY_DOCS_MAX_FILE_SIZE52428800 (50 MB)Maximum size of a single uploaded file
GAUZY_DOCS_MAX_EXTRACTED_CHARS5000000Cap on how much text is stored per document
GAUZY_DOCS_MAX_BINARY_BYTES10485760 (10 MB)Cap on a page's stored editing data
GAUZY_DOCS_QUEUE_CONCURRENCY2How many documents are processed at once per server
GAUZY_DOCS_STUCK_THRESHOLD_MINUTES30How long before a stalled document is picked up again

Storage and Versions​

VariableDefaultControls
GAUZY_DOCS_ORG_QUOTA_BYTES0Default storage quota per organization; 0 means unlimited
GAUZY_DOCS_VERSION_DEBOUNCE_MINUTES10Minimum minutes between automatic version snapshots while editing

A per-organization quota overrides the deployment default.

Inbound Email Capture (Advanced)​

Documents emailed to a per-organization address can land in the hub for review. This is off unless it is explicitly enabled and a webhook secret is set. Captured documents always go through review first and are never added to AI knowledge automatically.

VariableDefaultControls
GAUZY_DOCS_INBOUND_EMAIL_ENABLEDtrueMaster switch for inbound capture
GAUZY_DOCS_INBOUND_DOMAINunsetDomain used for capture addresses
GAUZY_DOCS_INBOUND_WEBHOOK_SECRETunsetShared secret used to sign inbound messages
GAUZY_DOCS_INBOUND_MAX_MESSAGE_BYTES26214400 (25 MB)Maximum size of one inbound message
note

The defaults above are the values that apply when a variable is left unset. The sample environment file shows illustrative values on some of these lines; because those lines are commented out, they are examples rather than the effective defaults.

Turning Documents On or Off​

The hub is governed by the Documents feature. Administrators enable or disable it per organization in Settings β†’ Features. When it is off, the sidebar entry disappears and the older Organization Documents and Help Center pages behave exactly as they always have β€” see Migrating from the Legacy Pages.