Antylogic
DemoApp login

399 operational guides

Docs sections
DocsPublic Docs
Device

Ask The Local Docs Assistant

Use the floating `Ask in Docs` panel on public Docs pages to ask a question and receive a browser-local answer grounded in selected public Docs articles. The assistant is read-only, needs no Antylogic account, and shows the articles it used when an answer is available.

In this article

Use this guide when a public Docs reader wants a quick answer without leaving the landing documentation experience. The assistant runs through the browser's built-in local AI and uses fetched public Docs article details as context.

The assistant does not create support tickets, change product data, or save a transcript. It selects up to four source articles, gives the current article priority when one is open, and displays the selected sources in the panel.

Before you start

  • Open a public Docs route rendered by `WikiScreen`: `/docs?lang=en`, `/manual?lang=en`, `/wiki?lang=en`, or an article route under one of those prefixes.
  • The public Docs index has loaded and contains at least one article; otherwise `Ask in Docs` is not shown.
  • `PUBLIC_API_BASE_URL` points to a backend serving `/api/public/docs`. Local landing builds default this to `http://localhost:8081`.
  • The browser supports `globalThis.LanguageModel` and reports an availability state other than `unsupported` or `unavailable`. If the model is `downloadable`, the first question can trigger a local model download.
  • For the GB English documentation experience, `en-GB` resolves to the implemented Docs locale `en`; there is no separate `en-GB` Docs corpus.

Instructions

  1. 1.

    Open Public Docs

    Open `/docs?lang=en`, `/manual?lang=en`, `/wiki?lang=en`, or a public article route. Wait for the Docs index to appear; the floating `Ask in Docs` button appears only when articles are available.

  2. 2.

    Open The Assistant

    Select `Ask in Docs` to open the local assistant panel. Read the readiness message before typing because it explains whether local AI is ready, needs a download, or cannot run in this browser.

  3. 3.

    Ask A Docs Question

    Type a specific, non-empty question about the public documentation, such as `How do I open the public booking page?`, then select `Ask`. A second question cannot be submitted while the assistant is responding.

  4. 4.

    Wait For The Local Answer

    The assistant ranks article summaries, fetches up to four article details from `/api/public/docs`, builds a prompt from those articles, and asks the browser-local model. If the model needs to download, progress can appear before the answer.

  5. 5.

    Review Source Articles

    Read `Local answer`, then check `Articles used` to see the selected source articles. Opening a source article is handled by the separate source-opening action.

  6. 6.

    Clear Or Leave The Session

    Use `Clear` to reset the question, answer, warning, source list, and download progress. Refreshing or navigating away also loses the transient assistant state.

Validations and limits

  • Blank questions are ignored, and `Ask` is disabled while availability is being checked, while a response is active, or when local AI is `unsupported` or `unavailable`.
  • No authenticated account, organisation membership, or location membership is required for `/api/public/docs`.
  • The assistant is read-only: it does not call a backend AI endpoint, persist a transcript, create support tickets, or mutate product data.
  • The current article is selected first when available, then the assistant fills the remaining context slots from ranked summaries, up to four articles total.
  • If article fetching, model creation, or prompting fails, the panel shows a warning instead of inventing an answer.
  • If the model returns no usable text, the panel shows `I could not find an answer in the available Docs context.`
  • Answers depend on the generated public Docs bundle. The analysed bundle reports `generated_at` 2026-06-18, while the documentation update context is from 2026-06-24.

Expected result

The reader receives a browser-local answer grounded in fetched public Docs article details and can review the source articles used. If the browser cannot run the local model, the panel still explains the limitation and Docs search remains available.

Related articles