AI chatbot for product documentation,
from the guides you already wrote
An AI chatbot for product documentation answers from your help center, setup guides, and feature reference, not from a generic model of other products. Heeya indexes the sources you add, retrieves matching passages, and ships as a script on the docs site you already have.
Turn your documentation into an AI agentFor technical writers, docs engineers, and product teams who already maintain a public docs site. Current plans are on Heeya pricing.
Docs assistant
Illustrative conversation
What an AI chatbot for product documentation does
Readers do not browse a 200-page help center the way writers outline it. They arrive with a concrete job: enable SSO, decode a 401, find the current rate limit, or check whether a toggle still exists. An AI chatbot for product documentation is a chat surface on that docs site that retrieves the matching article and replies in the reader's words.
Heeya's job is the retrieval layer plus the widget. You ingest the public docs, uploaded manuals, and optional question-and-answer pairs. You write guidance, test in the dashboard, then paste one script onto the HTML. That is not a docs host, not a search-and-replace for GitBook or Mintlify, and not a helpdesk.
If the buying question is "should a SaaS support team use a docs-grounded agent at all?", that lives on AI customer support for SaaS. This URL is narrower: turning documentation you already publish into a conversational assistant on that documentation.
The questions your docs already cover
These are retrieval jobs. They fail when the source is missing, stale, or mixed with a retired version.
Setup and configuration
Where a setting lives, which step comes first, required fields. Useful when a getting-started guide exists. Useless when the reader needs someone to inspect their workspace.
Feature reference
"Does this exist?", "What does this flag do?", "Where did this move?" The agent can retrieve the current article. It cannot demo the product UI.
Documented errors
Known messages and the fix you already wrote. Novel bugs, stack traces, and account-specific failures still need a person.
Limits and policy copy
Rate limits, plan differences, retention windows: only if that page is in the index. The agent will not change the reader's plan or usage.
Changelog questions
"Did we ship X?" works when release notes are ingested. It does not replace a status page or tell the reader which build they installed.
Hand-off when the article runs out
If you enable the form tool, the agent can show a contact form in the chat so the reader leaves details in one block. That is capture, not ticket routing.
How Heeya turns documentation into an agent
Five steps that match the product. Nothing here assumes a docs-platform plugin or a continuous git sync.
1. Sources
Add pages, files, or Q&A
Scrape selected URLs from a sitemap or llms.txt. Upload PDF, DOCX, PPTX, or TXT. Add question-and-answer pairs, including from a spreadsheet, for gaps that never became articles.
2. Guidance
Set tone on the docs site
Write system guidance. Set the agent name, welcome and popup copy, up to three suggested questions, bubble and chat colors, an optional logo, and English or French as the deployment language.
3. Test
Ask the questions in search logs
Use the dashboard chat against the same retrieval path before the widget goes live. Thin replies usually mean a missing page or a retired version still in the index.
4. Embed
Paste the script on the docs HTML
Copy the snippet from the connect page. There is also a full-page assistant URL if you want a standalone chat link next to the docs nav.
5. Review
Read transcripts and marks
Analytics lists conversations so you can open the transcript. Readers can mark a reply helpful or not helpful. Use that as a writing backlog, not as an automated quality score.
Choose how to ingest the docs
Pick the path that matches how the source already lives. For how to structure a corpus once it is in the index, use the knowledge base engineering guide. This page covers what Heeya will actually fetch.
Public HTML docs or help center
Paste the site URL. Discovery uses the sitemap, or falls back to an llms.txt index when that file exists. You then select which URLs to scrape. The crawler identifies itself as HeeyaBot, so docs owners who lock down bots can allow-list it.
When llms.txt lists Markdown exports, those .md sources are ingested as prose. That path is common on GitBook and Mintlify public sites. There is still no native plugin for those products.
Exported manuals and Q&A gaps
Upload PDF, Word (.docx), PowerPoint, or plain text. The website crawler skips PDF and DOCX attachments, so a "Download the PDF" link on the docs site will not appear in the index unless you upload the file.
Questions that never became articles belong as Q&A pairs. You can add them one by one or import a spreadsheet (.xlsx, .xls, or CSV).
Question to source map
Load the page that already answers the job. Skip secrets. Do not ingest two versions of the same guide.
| Reader asks | Source to ingest | The agent can | The agent cannot |
|---|---|---|---|
| How do I enable SSO? | Setup or SSO article | Restate the documented steps | Configure SSO in their workspace |
| Why am I getting a 401? | Authentication troubleshooting | Repeat documented causes and checks | Read their logs or tokens |
| Is feature X on my plan? | Current plan or limits page | Quote the loaded plan copy | See their billing record or change the plan |
| Does v2 still support Y? | The v2 guide only | Answer from that ingested version | Detect which SDK or app build they use |
| Where is the rate limit? | Limits or usage article | Point to that page if it was ingested | Inspect live usage or raise a quota |
Load first
- Getting started and authentication guides
- Feature reference that matches the current product
- Documented error pages readers already search for
- Plan, limits, and policy pages you want quoted
Add when answers are thin
- Changelog or "what shipped" notes
- Recurring tickets rewritten as Q&A pairs
- PDF manuals the crawler will not fetch
- A single current version of each guide
Do not ingest
- Private keys, tokens, or internal runbooks
- Drafts that contradict the public docs
- Retired versions that would still retrieve
- Ticket dumps, invoices, or customer PII
The embed is a script on the docs HTML
Heeya is a drop-in web widget. After you create an agent, the connect screen gives you a snippet in this shape:
<script async src="https://heeya.fr/agent/YOUR_AGENT_ID/embed.js" data-agent-name="Docs"></script>
Put it on any page that can load a third-party script, typically before the closing body tag. On Docusaurus, GitBook's site output, or a custom docs layout, that still means loading this script. There is no Heeya package to install and no props to pass for colors. Branding is configured on the agent.
If you want a page that is only the chat, the same agent also has a public assistant URL: https://heeya.fr/agent/YOUR_AGENT_ID/assistant.
That is the verified surface. There is no public Heeya programming interface for your backend, no native Slack or Discord channel, and no MCP server to claim here.
What retrieval changes, and what it does not
If there is chat history, the follow-up is rewritten into a standalone question. Similar chunks are retrieved from that agent's collection and injected as context. The model is told to use that context and to say when the information is not there. File names, including page URLs for scraped sources, are available to the model. Readers do not get a separate citation panel.
What this improves
- Answers can track the current ingested docs instead of a generic model's memory of other products.
- Updating a source, then re-scraping the page or replacing the file, is how you change an answer.
- When the chunk is missing, a well-behaved reply is that the information is not in the sources, plus a form if you enabled it.
- Unanswered questions in analytics are a writing list for the next article, not a score Heeya computes for you.
What this does not promise
- Retrieval reduces unsupported answers. It does not make them impossible.
- Stale, contradictory, or version-mixed docs produce stale, contradictory, or mixed replies.
- The agent does not execute account actions, inspect logs, or open pull requests against your docs repo.
- There is no certified "zero error" mode, and no SOC 2 badge to display on this page.
Fit, and not a fit
Heeya is a fit when
- The answers already live in a public help center, docs site, or exported manuals.
- You want a chat widget on that site without standing up chunking, embeddings, a vector store, and a chat UI yourself.
- A docs or product owner will choose sources, re-scrape after material changes, and skim conversations.
- A contact form in the chat is enough when the article does not cover the question.
Look elsewhere when
- You need the docs platform itself (authoring, versioning, preview). Heeya does not replace GitBook, Mintlify, or Docusaurus.
- You need a retrieval API, MCP server, or native Slack, Discord, or Teams app. Those are not available in Heeya today.
- You need continuous sync from git on every docs deploy. Refresh is a dashboard re-scrape of selected pages, or a file replace.
- You need the agent to take actions in the product, or to replace a helpdesk with assignment and live takeover.
Building the retrieval stack in-house means owning ingestion, chunking, embeddings, evaluation, and a chat front end as the docs move. Using Heeya means you operate the sources, the guidance, and the snippet, and you accept widget-only delivery. The cost and criteria for that decision are already written on the custom AI chatbot build vs buy guide.
Pricing
Heeya is sold as a subscription with a free starting tier. Plan names, message limits, and amounts change on the pricing page, so they are not copied here. Open current Heeya plans for the live grid.
Turn your documentation into an AI agentFAQ about AI chatbots for product documentation
Does this replace GitBook, Mintlify, or Docusaurus?
No. Heeya does not host, author, or version your documentation. It reads selected public pages and uploaded files, then answers from that index in a website widget. Keep your docs platform. Add the widget where readers already land.
Will the chatbot invent steps that are not in the documentation?
Retrieval-augmented generation reduces that risk by injecting retrieved passages into the prompt and instructing the model to say when the information is missing. It does not remove the risk. Test the questions your readers actually ask, and keep the ingested sources current.
Can I ingest a GitBook, Mintlify, or other docs-as-code site?
If the site is publicly crawlable, discover URLs from its sitemap or an llms.txt index and scrape the pages you select. When llms.txt points at Markdown exports, those .md files are ingested as prose rather than stripped HTML. There is no GitBook, Mintlify, or Docusaurus plugin. PDF attachments on the docs site are not fetched by the crawler; upload those files separately.
Do readers see a citation panel next to each answer?
No. Retrieved chunks include a file name, which for scraped pages is the page URL, and the model is instructed to include that URL in markdown when the information comes from a URL. The widget does not show a separate citation list. You review transcripts in analytics.
How do answers stay current when the docs change?
Heeya does not watch your repo or rebuild the index on every docs deploy. For website sources, you re-scrape selected pages from the dashboard and replace the indexed copy. For files, replace the uploaded document. Until you do that, answers follow the last ingested version.
Can the agent tell which product version the reader is on?
No. It answers from the sources you ingested. If v1 and v2 guides are both in the index, retrieval can mix them. Ingest the version you want quoted, and retire the rest. The agent cannot inspect the reader's installed SDK, app build, or plan.
Where are the current plans?
Current plans are listed on the pricing page. This page does not list amounts, because those can change. You can create a free Heeya account from the registration page.
Related reading
These URLs keep their own intents. This page stays the commercial docs-to-agent page.
SaaS support
AI customer support for SaaS
The cluster pillar for founders and CS teams. This URL stays the documentation product page.
Sources
Knowledge base engineering for AI chatbots
Inventory, hygiene, chunking, and drift: the pedagogical work this page does not repeat.
Decision
Custom AI chatbot: build vs buy
Engineering cost and maintenance when the alternative is owning the retrieval stack.
Put a conversational layer on the documentation you already maintain
Create an account, add the help center or guides readers already use, test the answers, then paste the embed script. That is the path Heeya supports today.
Turn your documentation into an AI agent