Skip to content

Knowledge Base

The Knowledge Base (KB) is your organization's permanent and curated knowledge asset within Queria. Unlike documents uploaded temporarily in conversations, KB documents are always available, prioritized in searches and shared across all authorized users.

What it is

The Knowledge Base represents your company's authoritative reference documents: operational manuals, internal procedures, policies, technical specifications, framework contracts and any other document that must always be accessible to the AI assistant.

KB documents have distinct characteristics:

  • Persistence: they stay available until explicitly removed.
  • Priority: in searches, KB results carry more weight than standard documents.
  • Sharing: accessible to all users in the organization (subject to configured permissions).
  • Curation: require special attention regarding quality, freshness and relevance.

Dual Cog-RAG: Knowledge Base vs User Documents

Queria's Cog-RAG architecture operates with a system called Dual Cog-RAG, distinguishing two document collections:

Knowledge Base (Permanent)

  • Company-wide shared collection.
  • Curated and validated documents.
  • Always available in all conversations.
  • Represents the "company truth" of reference.
  • Managed by the administrator or authorized users.

User Documents (Temporary)

  • Attachments uploaded by the user during a specific conversation.
  • Available only in the conversation where they were uploaded.
  • Automatically removed when the conversation closes (auto cleanup).
  • Ideal for ad-hoc analysis and quick comparisons.

How they interact

When you ask a question in chat, the assistant queries both collections:

  1. Searches the Knowledge Base for authoritative and consolidated information.
  2. Searches the user documents of the current conversation to add specific material.
  3. If enabled, also consults the certified external sources.
  4. Combines all results, prioritizing the KB, and generates a coherent answer with traceable citations.

This approach lets you always have access to consolidated company knowledge, enriched as needed by documents attached in conversation.

How to add documents to the KB

There are several ways to add documents to the KB:

From the Documents section

  1. Go to the Documents page from the navigation menu.
  2. Upload the file through any available method (upload, cloud, network, URL).
  3. During or after upload, select the "Add to Knowledge Base" option.
  4. The document will be processed and inserted in the KB collection.

From an existing document

  1. In the document list, find the document you wish to promote to KB.
  2. Open the document actions menu.
  3. Select "Move to Knowledge Base".
  4. Confirm. The document will be re-processed and inserted in the KB.

Bulk import

To add many documents to the KB at once, use the JSONL mode specifying the Knowledge Base flag per record, or upload files via cloud/network sync with the KB destination enabled.

When to use the KB vs Standard documents

ScenarioWhere to upload
Company operational manualKnowledge Base
Internal policies and proceduresKnowledge Base
Framework reference contractKnowledge Base
Technical sheet of a key productKnowledge Base
Company FAQsKnowledge Base
Draft contract to analyzeStandard or temporary document
Monthly report to consultStandard document
File received via email to verify on the flyTemporary document (chat attachment)
Active project documentStandard document
Consolidated internal regulationKnowledge Base

The general rule is simple: if a document represents an authoritative and lasting source for the organization, it belongs in the Knowledge Base. If it's an operational or temporary document, standard is enough.

KB management

Updating documents

When a KB document is updated (e.g. a new version of a procedure):

  1. Upload the new version of the file.
  2. The system detects the replacement and re-processes the document.
  3. Vector segments are automatically updated.
  4. Subsequent searches will return the updated information.

Removing from the KB

If a document is no longer pertinent or authoritative:

  1. Find the document in the Knowledge Base section.
  2. Select "Remove from Knowledge Base".
  3. You can choose to keep it as a standard document or delete it completely.

Monitoring

The Knowledge Base dashboard shows:

  • Total number of KB documents.
  • Processing status of each document.
  • Date of last update.
  • Coverage per topic (how many KB documents per category).

How the system searches in the KB

When the assistant receives a question, the KB search process follows these steps:

  1. Semantic analysis: the question is turned into a semantic vector.
  2. Hybrid search: the system runs both vector search (for meaning) and text search (for exact keywords) on the KB collection.
  3. Relevance and re-ranking: results are evaluated and reordered by pertinence.
  4. Integration: KB results are combined with those from other channels (standard documents, temporary documents, external sources).
  5. KB priority: in case of conflicting information, KB data gets higher weight.

Green badges in the answers indicate the information comes from the Knowledge Base.

Best practice

  1. Select carefully: don't put everything uploaded in the KB. KB quality depends on careful selection of authoritative documents.
  2. Keep the KB up to date: an outdated document in the KB is worse than a missing one. Review content periodically.
  3. Organize by topic: assign KB documents to relevant topics to ease targeted searches.
  4. Use authoritative sources: the KB should contain definitive and approved versions of documents, not drafts or temporary ones.
  5. Verify processing: after insertion, check that the document is in "Ready" state and segments are correct.
  6. Document changes: when you replace a KB document, inform affected users to avoid confusion on which version is in force.
  7. Don't overload: a KB with thousands of uncurated documents loses effectiveness. Better few quality documents than many approximate ones.

Queria v3.5.0 -- Document Intelligence platform

Queria - Document Intelligence con Cog-RAG