A chatbot on a documentation site is judged by a harder reader than one on a marketing page. Developers already know how to search. What they want from the assistant is the specific paragraph, a link that lands on it, and an honest “not in the docs” when it is not — because a fluent wrong answer about an API costs them an afternoon. So this comparison weights three things heavily: whether citations point at the section rather than the page, what the tool does when the docs are silent, and whether unanswered questions come back to you as a list you can write from.
We make one of these; it is listed first and is the least docs-specialised of the six, which we say plainly below. The other five are the tools that come up when docs teams compare. Prices and behaviours are what the vendors publish, and the closing section is a test you can run on any of them in an afternoon.
The comparison
| Tool | Self-serve entry price | Citations | When the docs are silent | Unanswered-question report |
|---|---|---|---|---|
| Matter Chat | $29/mo yearly: 2,000 replies, 200 pages | Every answer, linked to the page | Refuses and logs the question | Content gaps on Growth and above; unanswered questions on every plan |
| DocsBot | $49/mo: 5,000 credits, 5,000 pages, 1 bot, 1 user | “Sources” links, only the ones used | Classifies unanswered, escalates politely | Question clustering from Standard ($149) |
| kapa.ai | No public prices; demo-gated | Yes, with citations | Flags answer as “uncertain”, points to support | “Coverage Gaps” clusters, weekly to quarterly |
| Biel.ai | $50/mo: 500 pages, 2,000 interactions | Link to the specific page and section | Not documented; unanswered are logged | Content gaps, daily, from Professional ($150) |
| Docsie | $170/mo (Premium) on Docsie-hosted docs | Links to the exact page | Says it lacks the information, suggests support | Dashboard of searches without good answers |
| Inkeep | No self-serve cloud tier; enterprise | Rich citations in the UI | Admits it doesn't know, guides to support | Content & feature gap reports |
Matter Chat
Ours, and the honest framing first: Matter Chat is a grounded support assistant for any site, not a docs-as-code specialist. There is no Docusaurus plugin or GitHub connector; you point the crawler at the published docs and paste two lines into the generator's HTML template, which on Docusaurus, MkDocs or GitBook is a one-time edit. What it does well is the part this post weights: every answer carries a citation to the page it came from, the bot refuses when the docs do not cover the question rather than improvising, and the refusals come back as a list — content gaps, grouped, on Growth and above — so the docs improve because of the bot rather than in spite of it. Replies are metered with hard caps and no overage, and every plan includes team seats, which matters when the whole engineering team wants dashboard access.
Choose it if your docs are one site among several you support, you want a predictable bill, and section-level deep links are nice-to-have rather than essential. Choose one of the specialists below if you need GitHub issues, OpenAPI or Slack threads as sources, or a generator-native plugin.
DocsBot
The broadest connector list here — sitemaps, GitHub, Markdown, Notion, Confluence, GitBook, helpdesk tickets, Drive — and explicit citation controls in the widget. It is self-serve with published prices, which puts it ahead of half this list on its own. Model the credits: a standard answer is one credit on a base model and heavier models multiply that, up to 24×, and the bot stops with an error when the pool is spent. Unanswered-question clustering and escalation ticketing sit on higher tiers, so price the tier that has the feature you came for.
kapa.ai
The technical-docs specialist, with 23 connectors including GitHub issues and discussions, OpenAPI, Discord and Stack Overflow, answers with citations, an “uncertain” flag when grounding is weak, and a Coverage Gaps report that clusters the weak answers for you. One script tag installs it, with Docusaurus and GitBook plugins among many. The honest weakness is commercial rather than technical: there are no public prices and no self-serve — it is a demo and a sales conversation. For a funded developer-tools company that is fine; for a two-person team it is a wall.
Biel.ai
Built for docs-as-code teams and priced for them: $50 a month self-serve, a 14-day trial without a card, free for open source. Its citations go to the page and the section, and the install list is the longest of the six — a MkDocs plugin, Docusaurus, Sphinx, GitBook, Mintlify, ReadMe, VitePress, Starlight, Nextra, Antora and more. Two gaps to know: the content-gaps report starts on the $150 tier, and human handoff is not documented — the integrations are Zapier, Make, GitHub Actions and a Zendesk widget rather than a live inbox.
Docsie
A different shape: the chatbot is a feature of Docsie's documentation platform, answering only from docs you publish through Docsie. If you are already there, it is a toggle, it links to the exact page, and it says plainly when it lacks the information. If your docs live in Docusaurus or MkDocs, it is not a bolt-on option. The first tier with the chatbot is $170 a month and the metering is in AI credits.
Inkeep
The enterprise alternative to kapa, with a similarly deep source list — docs generators, GitHub, OpenAPI, GraphQL, Slack and Discord threads, helpdesks — rich citations, an explicit “admit when it doesn't know” stance, and an escalate-to-support tool that triggers on low confidence. Install guides cover Docusaurus, MkDocs, Mintlify, ReadMe and more. The open-source SDK is free; the managed cloud has no self-serve tier and no published prices, and the escalation tooling is enterprise-only.
Which one, for whom
- Docs-as-code team, self-serve budget, want section-level links and generator plugins: Biel.ai, then DocsBot.
- Developer-tools company with budget and GitHub, OpenAPI or Discord as first-class sources: kapa.ai or Inkeep.
- Docs already on Docsie: Docsie.
- Docs are one of several sites you support, you want cited answers and a fixed bill, and you can paste two lines into a template: Matter Chat.
The afternoon test
Every self-serve tool above has a trial; the demo-gated ones will usually run your content in a sandbox. Point each at the same docs section, then ask four questions: one the docs answer, the same one badly paraphrased, one the docs do not answer, and one that is out of date because you changed it last week. You are checking that the citation lands on the claim rather than the topic, that the paraphrase gets the same answer, that the out-of-scope question gets a refusal and not an invention, and that the stale one reflects the rewrite — which is a re-crawl question as much as a model one. The long version is how to test an AI support tool in twenty minutes, and writing help pages your bot can answer from is the fix for most of what the test finds.
