Chat Modes

Chat Mode is a workspace setting that controls how strictly RealTimeX stays grounded in workspace content when answering.

You can change it in Workspace Settings > Chat. The same concept also appears in Chat Embed as the embed Response mode.

The two modes

Chat

Chat is the broader, more flexible mode.

• answers general questions
• can respond even when the answer is not in your workspace documents
• works well for assistants, brainstorming, onboarding, and support flows where broader model knowledge is acceptable

Query

Query is the stricter, document-grounded mode.

• aims to answer from workspace context
• only answers when relevant context is found
• returns the workspace's Query Refusal Response when it cannot ground the answer

This makes Query the better choice for policy lookups, internal knowledge bases, technical documentation, and any workflow where you want RealTimeX to avoid improvising.

Where Query mode gets its answer

Query mode depends on the workspace's indexed content and retrieval settings. In practice, the answer quality is shaped by:

• the documents and notes in the workspace
• the Document Similarity Threshold in Workspace Settings > Vector
• the number of context snippets included from search
• the custom Query Refusal Response in Workspace Settings > Chat

If the workspace has no usable indexed content, or search cannot find relevant context, Query mode will refuse instead of guessing.

When to use each mode

Choose Chat when you want:

• a more conversational assistant
• help that can combine workspace context with broader model knowledge
• responses to general questions, even if the question is only loosely related to your documents

Choose Query when you want:

• answers grounded in workspace material
• fewer speculative or made-up answers
• a searchable knowledge-base experience instead of a general assistant

Common problems

Query mode says it cannot find relevant information

Usually one of these is true:

1. The workspace does not have indexed content yet.
2. The relevant wording in your documents is too different from the question.
3. The similarity threshold is too strict.

Practical checks:

1. Confirm the workspace actually contains processed documents, notes, or other indexed content.
2. Lower the Document Similarity Threshold in Workspace Settings > Vector if retrieval is too strict.
3. Ask with terms that more closely match the wording in the source material.

Chat mode answers questions that are not in the workspace

That is expected. Chat mode is intentionally permissive. If you want RealTimeX to answer only when the workspace can support the answer, switch the workspace or embed to Query.

Good defaults

• Start with Query for internal docs, SOPs, contracts, policies, and reference-heavy workspaces.
• Start with Chat for demos, customer-facing assistants, and exploratory conversations.
• If users complain that Query mode is too strict, adjust retrieval settings before switching modes entirely.