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

# Documentation assistant

> Prepare an agent that answers from reviewed product documentation, identifies unsupported questions, and links users to the right source.

A documentation assistant should help readers find and apply published instructions. Start with a narrow product area and a reviewed source set, then evaluate whether answers are accurate, current, and traceable.

<Note>
  You can prepare sources, agent configuration, and test definitions now. A saved knowledge attachment does not establish working retrieval, and automated test execution and conversation runtime remain subject to [current availability](/agents-platform/overview#current-availability).
</Note>

## Prepare the source set

<Steps>
  <Step title="Choose a bounded scope">
    Select the product version, audience, and tasks the agent should cover. For example, answer setup and configuration questions for one public integration. Exclude internal plans, customer records, and outdated pages from a public assistant's general knowledge.
  </Step>

  <Step title="Create and inspect knowledge documents">
    Use the [knowledge-base workflow](/agents-platform/configure/knowledge-base/overview) for reviewed URL, text, or file sources. Read stored content back and check processing status. Include stable source titles and canonical URLs so reviewers can trace instructions to their origin.
  </Step>

  <Step title="Attach sources to the intended agent">
    Create the agent first, then attach knowledge to the existing agent. The initial create-agent request rejects nonempty knowledge attachments. Review the selected branch and saved references before publishing a supported configuration.
  </Step>

  <Step title="Define answer boundaries">
    In the [prompt](/agents-platform/configure/agent-behavior/prompting), specify the supported product scope, how to ask for missing context, and when to say that the sources do not answer a question. Request source links only when the source is actually available. Do not let the agent invent package names, commands, settings, or release dates.
  </Step>
</Steps>

## Build a review set

| Question type | Expected result |
| - | - |
| Documented setup task | Correct prerequisites and steps, with a relevant source. |
| Version-sensitive behavior | Identify the applicable version or ask for it. |
| Missing information | Explain the gap and offer the appropriate support route. |
| Request outside scope | Decline to invent an answer or expose private material. |
| Conflicting sources | Identify the conflict for review instead of silently combining instructions. |

Store representative cases through [testing](/agents-platform/monitor/testing). Review them manually while automated execution is unavailable. Once retrieval and runtime are released, compare answers with the exact source snapshot and agent version used for each run.

## Maintain the assistant

Assign an owner to each source. When documentation changes, inspect the stored content and agent dependencies; do not assume a URL source automatically synchronizes. Recheck affected questions before treating an updated source as ready for public answers. Use [model guidance](/overview/models) to review the model selected for the task.

## Frequently asked questions

<AccordionGroup>
  <Accordion title="Does this enable an assistant on my documentation site?" id="documentation-assistant-faq-embed">No. This prepares the agent and source workflow. Website placement still requires a supported widget and session release.</Accordion>
  <Accordion title="Can I promise citations on every answer?" id="documentation-assistant-faq-citations">Only after testing the released retrieval and answer path. A prompt requesting citations does not prove that a returned link supports the answer.</Accordion>
</AccordionGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.