AI in Doxbrix is designed as product infrastructure, not as a single optional assistant panel. It supports the full documentation lifecycle: drafting, revising,...
AI in Doxbrix is designed as product infrastructure, not as a single optional assistant panel. It supports the full documentation lifecycle: drafting, revising,...
AI in Doxbrix is designed as product infrastructure, not as a single optional assistant panel. It supports the full documentation lifecycle: drafting, revising, reviewing, monitoring quality, answering reader questions, and governing usage at the workspace level.
The central idea is simple: documentation teams should be able to use AI without losing control over structure, accuracy, tone, or publishing standards.
AI in Doxbrix operates at four levels
| Level | Primary user | What AI helps with |
|---|---|---|
| **Page** | Authors | Drafting, rewriting, structure, and quality improvements while editing |
| **Project** | Documentation teams | Quality sweeps, issue discovery, and larger drafting tasks |
| **Workspace** | Leads and admins | Search, summarization, cross-project assistance, and policy enforcement |
| **Published site** | Readers | Grounded question answering from published documentation |
For authors
While a page is being written, AI helps authors move faster without breaking the editing workflow.
/ menu, have AI continue writing, generate a table from nearby content, or turn a description into a diagram.Writing assistance inside the editor
In the Block editor, AI is available directly where the author is already working. Instead of forcing writers into a separate chat flow for every task, Doxbrix exposes useful generation actions in context.
In the slash palette, the AI section includes actions such as:
| Command | What it does |
|---|---|
| **AI: Continue writing** | Drafts what comes next at the cursor. |
| **AI: Generate table** | Builds a table from the nearby content. |
| **AI: Generate diagram** | Turns a description into a Mermaid diagram. |
See Writing Copilot for the full set.
This design matters because it keeps authors focused on the document itself. AI is most effective here when it works as a structured editing tool, not as a disconnected brainstorming space.
Continuous quality analysis
Every page is analyzed while it is being edited. The Quality badge in the page header shows a live score from 0 to 100, and Document insights explains that score in actionable terms.
The analysis focuses on five dimensions:
- Readability — simplify sentences and tighten flow.
- Structure — fix headings and the section outline.
- SEO — sharpen the title, intro, and links.
- Accessibility — alt text and semantic structure.
- Completeness — fill gaps and add useful depth.
This makes AI useful even when you do not want generated text. Some teams use Doxbrix AI primarily as an editorial diagnostic layer rather than as a drafting engine.
Improve with AI
When a draft needs rewriting rather than just diagnosis, Improve with AI can revise the page in a guided way. Use it when you want to:
- tighten weak introductions
- restructure a page with poor heading flow
- add missing explanatory depth
- make technical instructions easier to follow
- address accessibility or SEO issues before publishing
Because it is connected to the quality model, it is more targeted than generic rewriting.
Beyond a single page
For larger jobs, the AI agent can draft pages and sweep a project for issues on a schedule and within approval guardrails, and Docs Health tracks quality and coverage across the whole project.
This is where Doxbrix moves from page-level assistance to program-level documentation operations.
For readers
Reader-side AI appears as the Ask AI assistant on the published site. It is designed to reduce time-to-answer while staying grounded in your approved documentation.
- It answers questions from your published content and cites the pages it used.
- It runs as a tool-using agent that searches your docs and reads pages to build a grounded answer.
- When it can't resolve a question, it can hand off to your support team with the conversation context — turning unanswered questions into deflected tickets.
Learn more in Site assistant and Support handoff.
The distinction between author-side and reader-side AI is important. Reader AI should not invent policy, workflow, or product behavior. Its job is to route people to the best answer already present in your published knowledge.
Response tiers
The assistant uses two tiers, chosen automatically by question complexity (and configurable by admins):
| Tier | Best for |
|---|---|
| **Spark** | Fast, instant answers for common questions. |
| **Atlas** | Deeper, multi-step reasoning for harder questions. |
This lets teams balance latency, depth, and cost across different AI surfaces.
You stay in control
AI in Doxbrix is governed at the workspace level. Admins decide which models are enabled, which features are available, how token budgets are managed, and how autonomous the agent is allowed to be.
That includes controls such as:
- model and tier availability
- feature-level enablement
- workspace token budgets and alerts
- agent autonomy and approval requirements
- terminology and style guardrails
- privacy and enterprise data controls
See Workspace AI policy for the full policy model.