Where Knowledge Base retrieval is used
Five AI surfaces call the same Knowledge Base retrieval engine: the Specialist agent’s search_documentation tool, the voice agent’s search_kb tool, Sidekick’s suggested answers, Self-Learning’s gap detection, and the search box on your public help center. Because they share the same retrieval foundation, an article that ranks well for one of these usually ranks well across the other surfaces too.
The five call sites
- Specialist agent’s
search_documentationtool — the AI agent inside a conversation looks up articles to answer a customer question. - Voice agent’s
search_kbtool — the same, but in a phone call. Limited to 5 results and short excerpts so the agent can speak the answer naturally. - Sidekick’s suggested-answer retrieval — when an agent is replying to a conversation, Sidekick offers relevant articles to insert.
- Self-Learning’s gap detection — when the system looks for cases where no good answer exists, it retrieves what it would have answered with.
- Public
POST /api/public/kb/:tenantId/search— the search box on your public help center, where market filters and language overlays apply.
Why this matters for authoring
The engine behind all five surfaces is the same hybrid keyword + vector retrieval approach, so optimizing an article for one surface generally improves it for every surface — including the human-facing help center search. Rankings should be broadly aligned, but they are not guaranteed to be identical everywhere. Some surfaces can have rollout- or latency-specific behavior: Specialist agents may use kb_retrieval_v2 when it is enabled for the tenant, while voice keeps the faster dedupe-first path. There is no separate authoring strategy per surface; the main per-surface differences are input limits (like the voice agent’s 5-result cap), rollout state, latency tradeoffs, and the market/language filters that apply on the public search.
For a precise account of how the engine ranks articles, see How the AI picks an article. For what to do about it as an author, see Write articles the AI retrieves well.