> ## Documentation Index
> Fetch the complete documentation index at: https://docs.interaone.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Ingest FAQs

> Author high-quality question and answer knowledge for direct semantic matching and normal agent retrieval.

FAQ sources are structured question/answer pairs. They can be retrieved by tools like other knowledge and also power a fast path that can answer without an LLM call.

## Write effective FAQs

<Steps>
  <Step title="Use a customer-shaped question">
    Write the phrase people actually ask, including the product or action when ambiguity is possible.
  </Step>

  <Step title="Make the answer self-contained">
    Include the direct answer, necessary constraints, and the next action. Avoid references such as “as above” that depend on page layout.
  </Step>

  <Step title="Separate distinct intents">
    Do not combine billing, security, and troubleshooting in one pair. Smaller intents produce cleaner semantic matches.
  </Step>

  <Step title="Index and test paraphrases">
    Ask the widget several natural variants and confirm the expected FAQ wins over similar entries.
  </Step>
</Steps>

## Fast-path behavior

For question-like messages, the agent embeds the text and searches the organization's FAQ vectors for one top result. A similarity score of `0.85` or higher and a non-empty stored answer cause the pipeline to publish that answer directly.

<Info>
  The threshold favors precision over recall because the response bypasses model reasoning. Lower-scoring FAQs can still contribute through normal tool-based retrieval.
</Info>

## Maintenance

* Review unanswered-question records for missing FAQ candidates.
* Consolidate near-duplicates that compete for the same query.
* Include dates or plan scope where policy changes over time.
* Reindex after edits and test both the direct answer and nearby negative cases.
* Remove answers that require real-time account data; use verified tools instead.