Set Up a Reverb 2.0 Target and AI Assistant

This topic connects a WebWorks Reverb 2.0 target to an AI Assistant on the WebWorks Platform. You create an Assistant on the Platform, point your target at it with the Assistant ID, generate a knowledge base archive from your documents, and upload that archive to the Assistant's Knowledge Base so it answers from your published help.

Before you begin: you need a confirmed WebWorks Platform account (see Create Your WebWorks Platform Account), ePublisher 2026.1, and a project that builds a WebWorks Reverb 2.0 target.

Overview

1. Create an Assistant on the Platform and copy its Assistant ID.

2. Turn on the AI Assistant in your Reverb 2.0 target and paste in the Assistant ID.

3. Generate the target to produce the knowledge base archive.

4. Upload the knowledge base archive to the Assistant.

5. Choose a model and reasoning effort, then publish and verify.

1. Create an Assistant on the Platform

1. Sign in to the Platform at https://auth.webworks.com.

2. Open AI Assistants in the sidebar and click + Create Assistant. Give it a name that identifies the documentation set it serves (for example, the product or help title).

3. On the Assistant's Configure page, its Assistant ID (a GUID such as 7d0897d3-086d-406b-915d-20e65766ae4a) appears at the top with a copy button. Copy it — you will paste it into your target in the next step.

4. While you are on the Configure page, give the Assistant its starting Instructions: paste the ready-made starter from A Starting Point You Can Paste and fill in its bracketed placeholders. Two minutes now means the Assistant answers in your product's voice from its very first question; you can refine the Instructions later as you test.

5. On the same Configure page, optionally give the Assistant an avatar image. Reverb 2.0 shows it wherever the Assistant appears in the chat — see Give the Assistant an Avatar below.

Keep one Assistant per documentation set (per Reverb target). The Assistant ID is what links a specific generated output to a specific Platform Assistant and its Knowledge Base.

Give the Assistant an Avatar

Since 2026.1, Reverb 2.0 renders the avatar configured for the Assistant on the Platform in every place the Assistant is represented in the chat: the welcome screen, each reply, the thinking indicator, the streaming reply, and the conversation-list header. Setting one puts your product's mark in front of readers instead of a generic icon.

Two things are worth knowing:

• The avatar travels with the Assistant, not with the output. There is no target setting for it and no need to regenerate: published output shows the current avatar the next time a reader opens the chat, so you can add or change it long after you publish.

• The chat falls back to the generic icon when the Assistant has no avatar, and also when an avatar image cannot be loaded — the avatar image must be reachable over http: or https:. The avatar slot is a fixed size in each position, so nothing on the page shifts while the image loads.

To style the avatar differently from the generic icon — for example, dropping the icon's background tint behind a photographic avatar — see Customizing the AI Assistant Avatar.

2. Turn On the AI Assistant in Your Target

In ePublisher, open Target Settings for your WebWorks Reverb 2.0 target and configure the AI Features group:

For the full reference description of each setting, see AI Features Settings in the Target Settings Reference and AI Features in the WebWorks Reverb 2.0 documentation.

A Single Build for Web and Local Use

With Hide AI Tab when file local enabled, the same generated output works in both places: served over http: or https: the Assistant tab is visible and functional; opened over file: (for example, bundled locally with a desktop product) the tab is hidden automatically. You do not need to maintain two separate builds.

3. Generate the Target

Build or generate your Reverb 2.0 target as usual. When generation finishes, ePublisher writes the knowledge base archive into the output directory next to the index.html entry point:

• A single project-wide archive named knowledge-<project-name>-<generation-hash>.zip, or

• One archive per parcel named knowledge-parcel-<parcel-name>-<generation-hash>.zip when Generate Knowledge Base Per Parcel is enabled.

The <generation-hash> changes on every build. If you automate uploads or deployment, match the archives by pattern (for example, files beginning with knowledge-) rather than by an exact filename.

4. Upload the Knowledge Base Archive to the Assistant

1. On the Platform, open your Assistant and scroll to Knowledge Base on the Configure tab.

2. Upload the knowledge-*.zip archive (or the per-parcel archives) by drag-and-drop or Click to browse. The Knowledge Base accepts .md, .pdf, .txt, and .zip files up to 512 MB each; each upload shows a status such as Completed.

3. When you republish, upload the new archive(s); to remove superseded files, select them in the Uploaded list and use the trash icon. With per-parcel archives you only need to re-upload the parcel whose content changed.

5. (Optional) Add Reasoning Fuel

Beyond the Grounding Knowledge it cites — the published help ePublisher generates, which the Assistant links back to through Landmark IDs — an Assistant can also draw on Reasoning Fuel (auxiliary knowledge): private, user-provided reference content that informs its answers but is not part of your published output and is not cited as a link. Use it to make the Assistant a strong troubleshooter for authoring, customization, build-automation, and "why" questions your help does not cover in depth. Upload your Reasoning Fuel files (clean, strictly structured Markdown is strongly recommended) to the same Knowledge Base as the grounding archive, and keep them accurate — they act like private weights applied on top of the model.

6. Choose a Model and Reasoning Effort

Both controls live on the Assistant's Configure tab. Use its Preview tab to compare models and reasoning efforts against your own content — the best way to find the right fit. A more advanced model is not automatically better: when your content is well written and well structured, a smaller model often answers just as well while being faster and cheaper.

• Model — for complex subjects that lean on Reasoning Fuel, use gpt-5.5; for simpler subjects, try gpt-5.4-mini. The gpt-4o models are fine for casual use but are not recommended for serious documentation Q&A.

• Reasoning — set to Off, Low, Medium, or High. Use Medium or High for multi-step troubleshooting; higher effort adds latency and cost (reasoning is billed as output tokens).

When a reasoning model is configured, the Reverb 2.0 chat shows a thinking indicator (and any streamed reasoning summary) while the model works, so readers get immediate feedback.

7. Publish and Verify

1. Publish the Reverb 2.0 output to a location served over http: or https:.

2. Open the output, click the AI Assistant button in the menu, and ask a question whose answer is in your documentation.

3. Confirm the Assistant responds and cites or links back to your published pages. If the tab is missing, check that the output is served over http:/https: (not file:) and that Generate Assistant is enabled with a valid Assistant ID.

The domain serving your published output must be listed under the Assistant's Origin Domains (on its Configure page) — enter the bare domain, with no http:///https:// or path. If it is not listed, the Assistant will not load even when everything else is configured correctly.

Next: Improve the Answers

Your Assistant now answers from your published help. To refine how it answers — tighter Instructions, well-curated Reasoning Fuel, and a repeatable way to test that changes actually help — continue with Optimize Your AI Assistant's Answers.