We use kapa as the 'Ask AI' assistant across both our developer docs and help center. At Xsolla, we provide the infrastructure for game payments and monetization, meaning developers using our docs need to get billing and commerce integrations right — fast. Kapa has genuinely leveled up that experience.

Instead digging through pages or chating with support team, developers get direct, cited answers from our own docs in seconds. The fact that it seamlessly handles multiple languages is a massive bonus for our global community. Bottom line: anything that keeps developers in the zone and unblocks them instantly is a huge win for us. Review collected by and hosted on G2.com.

Nothing major. The UI customization options for the widget could be a bit more flexible out of the box to perfectly match unique platform designs without much coding, though the core functionality completely makes up for it. Review collected by and hosted on G2.com.

Kapa trained an AI assistant on our entire documentation set (hundreds of pages) and lets users ask questions in plain language instead of hunting through docs. Answers come back with citations to the exact source pages, so people land on the right information fast. What's stood out beyond the core search: we plugged Kapa into our CLI via an MCP server, so developers can query our docs without ever leaving their dev environment, and we've built scoped chat experiences on individual doc pages. Sources refresh within minutes, so answers stay current with our releases.

The analytics have been genuinely useful too - we do a weekly review of the questions Kapa couldn't answer well to pinpoint exactly where our documentation has gaps. Setup was quick and the team has been responsive throughout. Review collected by and hosted on G2.com.

Nothing major. The analytics pages can get slow when you're reviewing a lot of sources at once. It's only minor, and the team has been quick to act on feedback - keen to see how the product keeps developing. Review collected by and hosted on G2.com.

We use kapa as an internal AI assistant for several of our engineering teams. It always seems to provide very accurate feedback nearly instantaneously. In the literally hundreds of questions that our engineering teams have asked it, I've only witness TWO hallucinations, and both were for a measly "Use this nonexistent Slack channel". The actual answer itself has literally always been accurate based on the knowledge that we've provided to kapa!

The UX is pretty stellar too with their Slack integration. It's the only way that we really utilize the app, and it works splendidly. The only grievance I have here is for Slack channels that utilize "forum mode" still require an @ mention of the bot to have it respond in Slack threads despite it responding without that in the main post. Fortunately, this is a very simple fix by ensuring that "forum mode" is not used.

With several different data source connectors available out of the box, we've yet to hit a place we have knowledge that we can't ingest. Slack channels, Confluence, Jira, GitHub... we have information everywhere and kapa does a great job pulling it all together into one Q&A spot!

With the answers being firmly grounded in our actual internal knowledge, and it operating strictly within the bounds of being a Q&A bot, the interaction is very consistent and dependable which is absolutely not the case with typical AI chatbots.

I find the pricing to be well worth the amount of time saved for our engineering teams. Their pricing model is very customer friendly and does not give any feeling that you so often see with "Well we're definitely getting ripped off in XYZ scenario if we use that part of the product." It's simple. It's clear. It's reasonable.

Customer support has also always been a pleasure to work with! They're exceptionally responsive and I've enjoyed the admittedly few interactions that I've had to have with them. Review collected by and hosted on G2.com.

I can really only think of the one thing - forum mode in Slack! I think it's fundamentally flawed and a rather intractable problem to solve to mesh forum mode with threaded replies. It's an easy fix to just not use it forum mode with your Slack integration. It really doesn't add much, if any, friction to the interaction. Requiring a user to @kapa.ai in Slack to invoke the chatbot is a tiny hurdle at worst in my opinion. Review collected by and hosted on G2.com.

We run kapa as the "Ask Al" assistant across our documentation, and the thing that matters most for us is accuracy on a genuinely complex docs surface. Harness is a modular platform - the same term can mean different things across products and kapa handles that with per-product source groups, so it doesn't blend answers across modules the way naive search does. When it isn't confident, it says so rather than inventing an answer, which for a developer audience is exactly the behavior you want. We also build our own Al in-house, so we seriously weighed doing this ourselves - but maintaining a retrieval pipeline well is close to a full-time job, and kapa dropped onto our existing docs as a script with no platform migration. The analytics showing where our docs come up short have been a useful bonus on top. Review collected by and hosted on G2.com.

Minor stuff. The traditional keyword-search experience wasn't fully at parity with what we had before on day one - particularly signposting which module a result belongs to. The Al answers were always our priority and have been excellent, and the team has been responsive on the search refinements since. Small, and steadily improving. Review collected by and hosted on G2.com.

The product just works, and the team behind it is genuinely great to work with. The analytics and insights built into the platform are something you don't fully appreciate until you think about losing them. Coverage gaps has become a core part of how we manage our docs sprint, and the custom auto-labels feature is exactly the kind of thing that makes it easy to track how different products are trending over time. Review collected by and hosted on G2.com.

Occasionally you'll see duplicate clusters in the coverage gaps suggestions, which can make it harder to prioritize what to work on. Review collected by and hosted on G2.com.

I find kapa.ai really helpful, especially for open source projects like Envoy (including Proxy, Gateway, and AI Gateway) where it's challenging to meet all user needs as usage grows. The pace of innovation at kapa.ai is amazing, from adding more information sources beyond docs to now including actual source code to the integrations in AI coding assistants, which is fantastic. The feature that provides statistical analysis of user questions helps identify missing features and clarify ambiguous ones, which unburdens the maintainer community. It's great that kapa.ai offers a safe space for people to ask questions, which can sometimes be intimidating in open source communities. The integration with platforms like Discord and Slack is cool because it helps people get assistance wherever they are, and it actually improves interactions within the community. The initial setup of kapa.ai was really easy and only took a few minutes. Review collected by and hosted on G2.com.

They have a really cool feature that allows you to segment your information into different products, but it takes a lot of manual configuration in the dashboard. I think it would be nice if Kapa had a programmatic way to configure the entire setup. They might have it, but I may not have access to it, I don't know. But right now, for me, I need to go and set that up manually. Review collected by and hosted on G2.com.

Our main customer-facing assistant pulls from our docs and our public roadmap through a web crawl, so users can easily get answers about current and upcoming capabilities. We've also used Kapa's retrieval to build a copilot inside our own product, so customers get answers grounded in that same content while they're actually using the tool, not just when they're on the docs site. On top of that, we run an internal version of Kapa for our own team, and it's quietly become the thing people reach for over our own internal search. Notion and Jira search inside our company are bad enough that people have said flat out they'd rather just ask Kapa, and it's especially handy when something is documented in more than one place since it finds the version with the right context instead of leaving us to guess. Review collected by and hosted on G2.com.

In some cases it is hard to cover specific edge cases without manual upkeep. Review collected by and hosted on G2.com.

We built our in-product AI agent on top of Kapa's MCP server, so it can search our knowledge base directly instead of us having to build that retrieval layer ourselves. It's become a core piece of how the agent answers technical questions, and when we've hit issues, the team has been quick to dig in and explain exactly what was happening.

Docs indexing is fast enough that we can push a fix to our content, wait a bit, and test again, which matters a lot when we're iterating on how well the agent handles a specific use case.

We've also been able to get fairly granular about what the agent does and doesn't reference, including steering it away from recommending features that are still in beta. Review collected by and hosted on G2.com.

We did run into some timeouts on the MCP connection that affected production traffic for a bit, but the team traced it to an infrastructure issue on their end and walked us through exactly what happened, which we appreciated.

Would be nice to have a bit more control over indexing speed and triggering re-syncs on demand, but it's a minor ask.

I also believe it sits on top of all of our user intents and could provide more insights into it, as well as guidelines into how to improve it (e.g., how to build "agent-ready docs", or which questions we should attend to).

Their dashboards are nice but missing a more AI/MCP native approach to the platform. Review collected by and hosted on G2.com.

I like how easy kapa.ai is to set up and configure. It's very trivial to connect new data sources, and it's very easy to link integrations within Zendesk, which is the CRM we use for support. It's just very easy to use. The reporting is pretty great too. It's easy to see how effective it's being and to spot what areas we can use to improve our documentation to make it even more effective. Kapa.ai helps us resolve issues before they even come to the support team, so a lot of the low effort tickets that would normally be sent to our team get deflected by the ticket deflector. It also gives our customers and our internal agents access to knowledge much faster. One interesting side effect is that onboarding takes about 50% less time, since our technical support engineers have access to all of our knowledge sources in one place and can quickly and easily ask questions. The initial setup was very easy. Review collected by and hosted on G2.com.

The reporting specifically for the ticket deflector 'lacks in comparison to some of the other features. I would really like to be able to go in and read the actual context of what's failing and what's not, like for our users. Review collected by and hosted on G2.com.

We have a large, fairly technical set of documentation, and what I wanted was an AI assistant trained on it without having to build and babysit a model myself. That's exactly what this is. You point it at your docs and it handles the hard part. Users can ask a question in plain language and get an answer back that cites the page it came from, which matters a lot for technical content, where a confident wrong answer is worse than no answer at all. Two things stood out for us. One is an "Ask AI" option right on the docs site, so people get an answer in context instead of searching around. The other is being able to put it inside our support workflow, so the team gets a drafted, sourced answer they can check before it goes out. It's trained on our own docs, including the content we pull in from our repos. Review collected by and hosted on G2.com.

The main thing to flag is that it didn't plug straight into the ticketing system our support already ran on, so we had to route questions through another channel to get answers in front of the team. It's workable, and something the team reacted to and has solved. Past that, the team was responsive and the setup matched what we actually needed. Review collected by and hosted on G2.com.