Glossary

Plain-language definitions of the terms used across ChatHQ and these docs.

Assistants and knowledge

• Assistant — An AI instance you configure for a specific role, such as support, sales or an internal helpdesk. Each assistant has its own instructions, knowledge base and settings. See Assistants.
• Instructions (prompt) — The text that tells an assistant what it is, how to behave and where its boundaries are. Sent to the AI with every message. See Instructions & behavior.
• Knowledge base / sources — The documents, web pages and FAQs an assistant answers from. Also called content sources. See Knowledge base.
• Content source — A single item in the knowledge base: one uploaded file, one web page, or one set of FAQs.
• Connector — A link between ChatHQ and another tool, such as Slack, Google Drive or SharePoint, so an assistant can read content from it instead of relying on manual uploads.
• Processing — The step after you add a source, where ChatHQ reads it and prepares it for retrieval. Usually completes within a few minutes.
• Draft / live version — Every assistant keeps numbered versions. The live version is what visitors talk to; the draft is the copy you edit before publishing. See Assistant versions.
• Publish — Make the draft the new live version.
• Restore — Bring back an earlier version of an assistant as the new draft.
• AI model — The underlying language model that generates an assistant's answers, chosen from the model catalogue. See Choice of AI model.
• Model catalogue — The list of AI models, grouped by provider, available to a workspace.

Answer quality

• Evidence verifier — The check that runs before every reply is sent, confirming each claim is supported by a passage in the knowledge base. See Evidence verifier.
• Grounded answer — An answer whose every claim is supported by a passage in the knowledge base.
• Hallucination — When an AI invents an answer that sounds right but is not supported by anything. ChatHQ's verification step exists to prevent this.
• Citation — The source passage an answer was built from, returned with every answer. See Source citations.
• Confidence score — A number between 0 and 1 returned with each answer, indicating how well the answer is supported by the sources.
• Fallback — The assistant's "I don't have that information" reply. Good when the answer truly is not in the documents; a problem when the answer was there and it was missed. The wording is yours to set.
• Fallback message — The specific text shown to visitors when the assistant falls back. Configured under the assistant's Fallback & Privacy settings.
• Service disruption — A failure such as a provider timeout, shown to visitors as a neutral notice rather than the fallback message. See Failure visibility.
• Prompt injection — A message crafted to trick an AI into ignoring its rules, revealing its instructions or taking on a different role. See Resisting manipulation.
• Guardrail — A safety check that can stop a request before an answer is generated.
• Agent / agentic — An AI that takes actions (sends, books, buys) instead of only answering. ChatHQ assistants answer; they do not take actions on your systems. See No fake actions.

Delivery

• Widget — The chat interface you add to your website with a small JavaScript snippet. See Widget overview.
• Embed — Putting the assistant on a website by adding the widget snippet to the page. See Embed code.
• Publishable key — A key designed for client-side use. It identifies your workspace in the widget snippet and in API calls, and is safe to include in a web page. See Authentication.
• Allowed domains — The list of websites the widget may load on. Requests from other domains are rejected. See Allowed domains.
• REST API — Server-to-server access to assistants and conversations. See REST API.
• Session — One conversation. Reusing the same session id across messages lets the assistant keep the context of earlier turns.
• Rate limit — A cap on how many requests can be made in a period, applied per IP address and per workspace plan. See Rate limits.

Platform

• Workspace — Your main container. It holds assistants, widget configuration, publishable keys, members and settings. See Workspace basics.
• Member roles — What each person in a workspace can do: Owner, Admin or Member. See Members & roles.
• Plan — The subscription that sets a workspace's usage limits and features. See Plans & billing.
• Analytics — Conversation, message and engagement figures for your assistants. See Analytics dashboard.