Jamdesk Documentation logo

AI Chat

Every Jamdesk docs site includes a built-in AI chat assistant. Visitors ask questions and get answers with citations from your documentation.

Every Jamdesk site includes a chat assistant that answers visitor questions from your documentation. It retrieves relevant sections, sends them to Claude, and streams a response with links back to the source pages. Chat is enabled by default on all plans at no additional cost.

Here's the chat assistant in action:

Use Cases

A developer integrating your API asks "How do I authenticate requests?" and gets back the exact steps with a code example, plus a link to the source page.

How It Works

1
Build indexes your docs

During each build, Jamdesk splits your pages into sections and stores them as searchable embeddings.

2
Visitor asks a question

The "Ask AI" button (or Cmd+I / Ctrl+I) opens a chat panel on any page.

3
AI answers from your docs

Relevant sections are retrieved, sent to Claude as context, and the response streams back with citation links to the source pages.

Features

FeatureDetail
Streaming responsesAnswers arrive in real-time
Citation linksEach answer links to the docs pages it drew from
DisambiguationWhen a question matches multiple topics, the AI asks which one the visitor means
Conversation context10-message history, persisted per browser tab
Starter questionsAuto-generated from your docs, or set your own in docs.json
Keyboard shortcutsCmd+I / Ctrl+I to toggle, Escape to close
Markdown renderingCode blocks, tables, and lists render inside responses

Chat panel empty state with starter questions

Configuration

Chat works out of the box with no configuration required. Add a chat block to your docs.json only when you want to customize the starter questions or turn chat off. The chat field is optional, and both of its settings have sensible defaults.

docs.json
{
  "chat": {
    "enabled": true,
    "starterQuestions": [
      "How do I get started?",
      "What API endpoints are available?"
    ]
  }
}
FieldTypeDefaultDescription
enabledbooleantrueSet false to remove the Ask AI button, chat panel, and keyboard shortcut from your site
starterQuestionsstring[]auto-generatedUp to 4 questions shown when the panel opens (5-200 characters each). Auto-generated from your docs when omitted; set to [] to show none

Starter questions

Starter questions are the suggested prompts shown in the chat panel before a visitor types anything: one-click examples that hint at what they can ask. When you leave starterQuestions out of your docs.json, Jamdesk auto-generates them from your documentation on every build, so they stay relevant as your docs grow.

To set your own, list them in the starterQuestions array. This is handy when your support team fields the same questions over and over and you want to surface those first:

docs.json
{
  "chat": {
    "starterQuestions": [
      "How do notifications work?",
      "Tell me more about analytics"
    ]
  }
}

You can set up to 4 starter questions, each between 5 and 200 characters. To show no starter questions at all, set starterQuestions to an empty array ([]).

Disabling chat

Chat is enabled by default on every plan. To turn it off, set enabled to false. This removes the Ask AI button, the chat panel, and the Cmd+I / Ctrl+I shortcut:

docs.json
{
  "chat": {
    "enabled": false
  }
}

Disabling chat doesn't affect search. AI chat and search are separate features, so your docs stay fully searchable even with chat turned off.

Limits

LimitValue
Max message length2,000 characters
Max response length2,048 tokens
Rate limit10 requests / 60 seconds per visitor per site
Conversation history10 messages per tab
Starter questions4 max

Chat works on custom domains. The /_chat endpoint is same-origin, so it functions on any domain where your docs are served.

During each build, Jamdesk splits your pages into sections and stores vector embeddings for each one. When a visitor asks a question, the system runs a hybrid search (keyword matching combined with semantic similarity) to find the most relevant sections across your docs. Those sections are sent to Claude as context with instructions to answer only from the provided documentation. Citations are extracted by matching page references in Claude's response against the retrieved sections.

What's Next?

llms.txt

Auto-generated page index for AI tools

MCP Server

Let AI tools search and query your docs programmatically

docs.json Reference

All configuration fields including chat settings

How Jamdesk Works

The build pipeline that powers chat indexing