Workspace
The top-level tenant. A workspace owns billing, members and roles, audit logs, usage metering, SSO connections, bot integrations, and helpdesk integrations. Workspace owners manage everything; members get access based on knowledge base permissions.Knowledge Base
Typically one product’s docs, a repo, or an internal wiki. Knowledge bases own sources, client keys, agents, actions, analytics, and per-knowledge-base LLM settings (persona, tone, custom instructions, provider/model overrides).Source
A connected content origin inside a knowledge base. Supported types arewebsite, github, github_issues, github_pull_requests, github_discussions, notion, confluence, gdrive, onedrive, tickets, files, slack, discord, zendesk, jira, jira_service_management, linear, salesforce, salesforce_cases, openapi, discourse, stackoverflow, youtube, s3 and qna. Each source re-syncs on a schedule, on demand, or via webhook (git push). See Knowledge Sources.
Source Group
A named collection of sources within a knowledge base — use groups to model docs versions (v1, v2, latest), separate products, or internal-only content. Client keys can be restricted to specific groups (allowed_groups), and chat/search requests can pass group_ids to narrow retrieval further. See Source Groups & Sync.
Crawl / Sync
An ingestion run. A crawl fetches content from a source, normalizes it, and (re)indexes changed pages using content-hash change detection. Every source type runs on a fixed cadence of its own — 5 minutes for Notion, 24 hours for a website crawl — with no schedule to configure. Sync runs one on demand for any type with an upstream, and git sources reindex on an inbound webhook.Document
A normalized markdown page with metadata (URL, title, headings, crawl timestamp). Documents have a visibility flag:public documents are retrievable by widget/SDK/bot traffic; restricted documents are held back from the widget, SDK, public API, MCP and A2A, and stay available to authenticated dashboard and Ask users — and to the Slack/Discord/Teams bots, which answer inside your workspace.
Chunk
The retrieval unit: a heading-aware segment of a document stored with both a vector embedding (for semantic search) and a full-text index (for keyword search). Answers are built from the top-ranked chunks after Reciprocal Rank Fusion.Conversation, Message, Citation, Feedback
Each chat exchange is persisted as a conversation of messages. Assistant messages carry citations — numbered[n] markers mapped to source documents — and can receive feedback ratings that feed the knowledge-gap analytics and retrieval boosting.
Client Keys
Per-knowledge-base integration credentials:
Public keys support origin allowlists and optional reCAPTCHA enforcement; secrets are hashed at rest and shown once. See Authentication.