And we're live! Check out the @TechCrunch article here: techcrunch.com/2024/10/03/ho… :)

Spent the weekend with 100+ founders (incl. @sama) at @ycombinator's AI retreat. Here are my 5 biggest takeaways for 2025: 1/ o3 isn't the story: Everyone's hyped about OpenAI's o3, but o3-mini is the real game-changer. It brings next-level reasoning capabilities at a fraction of the cost. Releases in a few weeks. 2/ The "better than human" paradox: AI clearly outperforms humans on specific tasks, yet struggles with adoption. Why? Users expect perfection from AI but accept human error as normal. 3/ Fine-tuning's limited role: Most production use cases are moving to custom RAG pipelines. Fine-tuning often leads to alignment issues and higher hallucination rates in real-world deployments. 4/ Voice AI is truly here: One YC startup (@happyrobot) has already automated 100k+ high-value freight calls. While the demos look cool, doing this well requires very hard engineering. 5/ The eval secret: Most are building in the dark. Best kept secret in AI? Every startup is building their own eval suite because what works for fintech fails for consumer support. Being part of YC continues to be the biggest unfair advantage in tech. Can't recommend applying to Spring 2025 batch enough. Huge thanks to our incredible partner @sdianahu and @garrytan for the invite. The caliber of founders in the room was unreal. 🚀

launch alert: gpt-4 powered docs widget for @nextjs we gave @kapa_ai access to all: - @nextjs docs + tutorials + templates + blogs - @vercel docs + guides demo: kapa-widget.vercel.app/widge… 🕹️ what framework should we do next?

This is your technical support on @kapa_ai. Our customer deployed in January, and they saw a ~30% reduction in support tickets. This isn’t from making support harder to get to, or impossible to log a ticket with, or anything like that. It’s simply adding an AI assistant that answers questions with real answers, grounded in documentation. An AI assistant that can combine answers, and knows when to say “I don’t know.” An AI assistant that’s actually enjoyable to use. 😅 Now, they have more time to improve their docs, focus on larger tickets, and improve their DevEx. This is why we do it!

Most people don’t realize @ycombinator doesn’t end with demo day. Now, 1.5 years after this picture was taken…a few post-batch thoughts: (🧵)

launch alert: gpt-4 powered docs widget for @Shopify we gave @kapa_ai access to all @ShopifyDevs: - apps, themes, storefront, and marketplace docs 📒 - full API reference 🌐 demo: kapa-widget.vercel.app/widge… 🕹️ what framework should we do next?

excited to announce that we're part of @ycombinator with @kapa_ai and have moved to San Francisco! 👋

our latest blog post on ai hallucinations just hit HN front page 🎉 go read it here: kapa.ai/blog/ai-hallucinatio…

“It ingests a lot of the developer documentation, the YouTube videos, the chat history…(kapa) just keeps getting better and better” Thanks @ycombinator for the shoutout, and for highlighting one of my favorite parts about kapa. If we’re building the AI assistant layer for your documentation, we want to be comprehensive. So it’s not just articles - it’s YouTube videos, chat transcripts, tutorials, and emails. All of that goes into making kapa the most effective support layer for your product. Your support teams aren’t running on empty, or 5 tickets behind on turning that sev-3 into a knowledge base article. They can shift focus - to analyzing user feedback, FAQs, and requests…building out more helpful documentation. WHICH…kapa uses to improve the answers to all the questions your new users have. It’s an impressive flywheel, and I’m proud to see how much it’s caught on.

“(kapa) will say when it can't find information, instead of making shit up” I love that happy customers (like $2.45B semiconductor giant Nordic Semiconductor) get positive feedback like this, even when their users don’t get the information they’re looking for. Nordic knows what so many companies haven’t found out yet: an AI assistant that makes things up is next to useless. (if you’re actually trying to help your users, at least.) Your customers won’t trust your docs, and they’ll leave frustrated, confused, without answers. kapa gives better answers, AND knows its limits: - Only sourcing information from verified company docs - Leaving citations and links throughout its answer - MOST IMPORTANTLY, knowing when to say “I don’t know” That’s how you create an AI assistant that actually improves your documentation. Proud to partner with @NordicTweets, and see these posts in the wild.

AI news: @Docker just announced their in-app AI agent Gordon. And they're using @kapa_ai to power it. 🐳 Here's three reasons why this matters: 1. Deep platform integration > chat interfaces. Gordon lives right in Docker Desktop and CLI where devs already work (used by 17 million monthly devs!) 2. Smart engineering resource allocation. Docker team focuses on platform while kapa handles the core RAG/LLM knowledge layer. 3. Faster time to market. Leveraging existing infrastructure lets Docker ship production-ready AI agent features as some of the first. I encourage y'all to read their technical deep-dive in their newsletter. I think we'll see a lot more companies follow this approach in the coming months.

we also use @Docker at @kapa_ai, and it's very good too :)

Why do AI models make stuff up? After working with 100+ companies including @Docker and @Reddit we wrote up a brief literature review of what works and what doesn't when it comes to reducing LLM hallucinations. Link: kapa.ai/blog/ai-hallucinatio…

“Why do LLMs suck for embedded systems?” is a question I see come up a lot. Funny enough, the counter-example always seems to be a kapa customer, Nordic Semiconductor. I’m not surprised. Their customer-facing AI assistant on their public docs does a few things extraordinarily well: - It pulls from Nordic’s rich knowledge base, which is very actively maintained by the Nordic team - It lists citations for its answers, so users have confidence, and can read the docs if you want - It knows when to say “I don’t know”, which is IMO the most critical thing an AI support system should do. Every time. It boggles my mind how few companies get that last point right. It’s like they’re terrified they won’t be taken seriously because their AI isn’t some all-knowing oracle. We think the opposite, and so do customers. You take an LLM MORE seriously when it’s honest about its blind spots. That way, you know the answers its giving are cited and grounded in reality. Proud of our work with @NordicTweets and our 100+ other customers who are doing this right.

Can reasoning models like o3-mini improve state-of-the-art RAG pipelines? We ran 58 different evaluations, and the results were...mixed. Despite recent model advancements, we ran into the "reasoning ≠ experience fallacy" for tool calling. Let me explain. 1/

testing @kapa_ai on new internal company data formats and receiving customer feedback like this 👇

I’m so honored and proud to see Kapa listed on EU-startups’ “10 promising Danish startups to watch in 2025”. Denmark has the highest number of unicorns per capita in Europe, and created billion-dollar legends like @Unity, @Zendesk, and @Pleo. This is not a coincidence; it’s a testament to the concentration of talent and tenacity I’ve seen and worked with. I wholeheartedly reject the idea that Europe can’t innovate like any other part of the world, even tech hubs like New York and SF. (And yes, we’re steeped in Silicon Valley culture, but that doesn’t mean we’re forgetting our roots.) There are incredible entrepreneurs and innovators in Denmark, and I’m honored we’re on this list.

insanely excited to finally announce this!

launch alert: gpt-4 powered docs widget for @Modular_AI we gave @kapa_ai access to all: - developer documentation + blog articles - open github issues + bugs + discussions demo: demo.kapa.ai/widget/modular 🕹️ hoping this makes picking up @ChrisLattner + team's mojo lang easier

the @ycombinator HackerNews front page jobs perk is super underrated as a founder. check out our new role here if if you get excited about applied RAG research @kapa_ai: ycombinator.com/companies/ka…

demo alert: gpt-4 powered docs widget for @webflow we gave @kapa_ai access to: - 1,300 pages of webflow documentation - 8,500+ community forum answers demo: demo.kapa.ai/widget/webflow 🕹️ hoping this makes it easier to build @webflow sites!

6 spots away from HN front page. so dang close.🤞 news.ycombinator.com/?p=2 (pls help)

Thanks for the @kapa_ai shoutout @sdianahu in the latest @ycombinator podcast on how vertical AI companies can create the next 300 unicorns. 🦄

🚀 Exciting news: @kapa_ai is now powering AI assistance across @Reddit developer documentation! We've integrated our AI assistant directly into the @Reddit Developer Platform (Devvit), helping developers build everything from word games to community tools. Three reasons why this matters: 1. Massive platform evolution: Reddit is transforming subreddits into revenue-generating platforms for developers - from marketplace tools to community bots (straight from their S-1 filing) 2. Perfect timing: Reddit just launched a $116k Games Hackathon, with prizes for building word games, puzzles, and tabletop experiences 3. Developer-first approach: Every question about Reddit's platform can now be answered instantly with kapa.ai's help - making development faster and more accessible Want to see what we built together? (links in comments) 👉 Try the "Ask AI" feature directly on their docs 👉 Join the Games Hackathon before Dec 18th 👉 Check out the Reddit guide for kapa-powered Ask AI Huge thanks to the @Reddit team for trusting @kapa_ai. This is just the beginning!

launch alert: gpt-4 powered support AI for @Roblox game devs we gave @kapa_ai access to all @robloxdevrel's: - dev documentation and API reference 📝 - discourse forum posts ✉️ - education materials 🎒 demo: demo.kapa.ai/widget/roblox 🕹️ what framework should we do next?

My dad is a hardware engineer and has built with Nordic Semiconductor chips his entire career. So the fact that @NordicTweets now also use @kapa_ai to power "Ask AI" on their docs is very wholesome. (proof pic from recent family visit)

My first conference talk is live! 🎙️ Thanks @APItheDocs for organizing the "AI The Docs" on LLM best practices for technical documentation. Learn how companies like @Docker @mixpanel @temporalio @CircleCI are all adopting LLMs like @kapa_ai trained on their docs to improve developer experience. In this talk I dive into how to structure documentation for LLMs and wanted to share some best practices to consider for others considering this approach.

50% faster technical support replies. 1 API call. 🔌 The smart folks at @RudderStack built a @kapa_ai API-powered support workflow to automate ~70% of common questions their technical account managers have to answer.

We're seeing a new trend: AI-forward companies like @Mapbox , @mondaydotcom, and @CircleCI are deploying LLM assistants in their products. I analyzed their results and approach in a detailed step-by-step breakdown. The findings are impressive: - Instant, contextual support without leaving the product - Up to 20% reduction in support tickets (Mapbox's case) - Continuous improvement of documentation from user queries Want to learn how it works and how you could implement it? Check out the link below.

so cool to scroll through twitter and see @kapa_ai 💜 in the wild thanks @jamckernan!

Wow. @Docker just announced their in-app AI agent Gordon. And they're using @kapa_ai to power it. 🐳

some more @kapa_ai love in the wild ft. @roboflow 🦝

a bit of @kapa_ai love spotted in the wild on HN 😍

Easy. @oneseventwonine on DAO 101: 1729.com/daos.

we need @bobvanluijt to appoint @CShorten30 as Chief Weaviate Professor with content like this🎉

launch alert: gpt-4 powered docs widget for @twilio we gave @kapa_ai access to +4,000 pages of docs incl. sms, voice, CLI, conversations and IoT 🧠 demo: demo.kapa.ai/widget/twilio 🕹️ hoping this makes navigating docs for @TwilioDevs easier!

If you’re building a technical product, sometimes the best product partnerships are two-way. Some of our favorite products are heavy Kapa users: - Next.js (website) - OpenAI (LLM calls) - Weaviate (vector database) - Reddit, Inc. (coffee breaks :D) - Sentry (application monitoring) - Docker, Inc (local development) - PaperCut Software (office printer software!) If you’re building a technical product, ask your partners about what tool they’re using to solve the problem you’re solving. Worst-case, you learn a bit about why they picked that tool. Best-case, new customer. If everyone can benefit…why wait?

So you want to build an AI startup and get backed by @ycombinator in 2025. Some tips: 1/

Not surpised. We're lucky enough to count @josephofiowa as an angel investor in @kapa_ai and he's consistently my first text and the most helpful sparring partner.

Uhh what tool?

so freaking cool to see feedback like this of @kapa_ai in the wild.

to see if models like o3-mini, deepseek r1 and grok 3 work for RAG pipelines we ran +50 evaluations so it's awesome to see the results are getting frontpage hacker news love news.ycombinator.com

Sentry scaled to over $100mm in ARR, 100k+ customers, and $3 billion in valuation. Now, they’re our newest customer, with kapa’s “Ask AI” powering their documentation. Three reasons this is big: 1. Millions of developers are now browsing AI-first documentation. They’re getting used to saving time, getting accurate answers, and giving direct feedback. This is how documentation should be. 2. Easy shortcut right on the docs homepage. ⌘+k gets you instant access to answers. Check out the video below - easy access to answers in natural language. 3. A wide breadth of support for the top-N most popular frameworks. Sentry’s team doesn’t need to spend time building out FAQs for each individual framework - they let kapa do the heavy lifting. And best of all, we’re happy customers of their Application Monitoring product. The best partnerships are two-way after all. Go try it for yourself at docs dot sentry dot io!

congrats to my fellow danes and @ycombinator s23 batchmates for landing some super impressive early customers (incl. NYT!) and a great round 🇩🇰🎉

Wow. kapa.ai just took the stage at Mobile World Congress 2025 (attended by 100k technology leaders!). Nicolas Grenié (Developer Advocate at @typeform) gave an epic talk: "Developers are the new buyers: why developer experience matters." Here are three key insights he shared: 1. Optimize for TTFHW (Time To First Hello World): Great developer experiences minimize the time it takes a developer to get started. 2. DX is much more than docs: It spans Discord/Slack communities, forums, API playgrounds, and making yourself available to new developers. 3. AI-powered chat is changing the game: AI-driven documentation agents, like @kapa_ai, proactively deliver clear answers, increasing developer trust and satisfaction. Thanks for the shoutout, Nico! Link to the full presentation (highly recommended) in comments.

proposal to change your landing page hero 👇

30% of our company is ex-McKinsey, including myself. We took the best lessons from working there, and applied them to building a YC-backed startup. Steal our playbook: (Before you get any ideas, 30% means 3 people - we’re a lean operation) Alright, those lessons: - You have to actually solve a real problem. If you can’t make a business case that appeals to an exec with your product, you’re nothing but a cool tool. Simple as that. I think this is especially true in AI startup land, since it’s so easy to just make your pitch “we do X but with AI”. That might win you a few early deals, but trust me, those customers won’t stick if you don’t solve an actual problem. - Every enterprise deal is like its own mini consulting project. If you don’t have all the right people in the room from day one, you are wasting your time. You have to manage the process, timelines etc - don’t expect anyone else to. Find this out as early as possible in the sales cycle. - Write a good top down email. It’s way clearer. Concise communication is becoming a lost art, and it’s sad. McKinsey had this concept called the “Pyramid Principle” that works well for us - at a high-level, you start your communication with the answer, then provide supporting arguments, then data. Everyone can understand your message quicker. - Slides still work. That’s it, that’s the bullet point. (But you can probably get away with 5 slides instead of 500 😅) Now that I think about it, we’ve all realized you also have to UN-learn a lot of stuff from McKinsey. In startups, 99% of what matters is doing, and not overthinking what to do. In strategic consulting projects, it’s the exact opposite. That’s a mindset switch you have make very quickly; otherwise, you won’t get any traction. All this to say, I’m proud of our roots. There’s a reason McKinsey got where they are. We learned a lot there, and their processes work. Having a repeatable process for success helped us get where we are, too.

The best product stories aren't told by founders - they're told by customers. So when a $4B semiconductor leader with 1,770+ patents tells your story, you listen. Check out @siliconlabs showcase how they're using @kapa_ai to help millions of developers by giving an AI assistant access to all of their technical materials and making it available to their users. I'm honestly blown away by how they made a way better demo video than we could have. 😅 Head over to the Silicon Labs website and look for the "Ask AI" assistant.

want to come hear the funnest of conference talks? ✨ I'll be speaking at the @APItheDocs special one-off conference on AI best practices for developer docs next week!

WOW. The most-starred Apache project (Superset) just launched Ask AI chat on their docs, and the effect was immediate: Less than 10 minutes after using kapa, a developer fixed an issue that had been plaguing them for days. 1. They announced the launch in their Slack community 2. A user saw the announcement 3. 10 minutes later, their issue was fixed This is the type of developer experience that’s only possible when you have an AI assistant positioned closely to your docs. Because Superset’s has access to everything it needs to help its users: 1. All documentation 2. Github open issues 3. PRs 4. READMEs So any error known error message or situation is covered by the AI assistant. It can provide deep, personalized help beyond “have you tried turning it off and on again” or “works on my machine”. Love seeing instant success deployments like this. Plus, I’m just so excited to think about how many developer headaches this will fix. Apache is the biggest software nonprofit in the world, home to huge, famous projects like .@ApacheKafka and .@ApacheSpark. Superset being their most-starred project means we’re helping so many developers. If this is what happens within a few minutes of .@ApacheSuperset's deployment, I can’t wait to see how many issues it helps resolve for developers in the months and years to come. Another happy .@kapa_ai user.

Exciting news: kapa is powering the AI assistant for $4 billion semiconductor giant Silicon Labs 🚀 Three reasons why this matters: - Easy to talk to: Kapa understands natural language, making the interactions with their 10k documents much easier - Comprehensive view: Since kapa maps out the entire knowledge base, it can suggest relevant articles and resources without making their users find them - Platform-agnostic: Kapa works easily on any device, meeting Silicon Labs’ users where they actually are I’m proud of this deployment. @SiliconLabs has 10,000 pages of documentation, with links to 2500 PDFs, 120+ GitHub repositories, and nearly 400 YouTube videos. If you’re in search of answers, you can’t sift through all that manually or easily. But thanks to @kapa_ai, it’s all easy. API docs, release notes, user guides, schematics…whatever you need, we’re here to help. Check it out for yourself!

ngl, but pretty cool to see the @mondaydotcom enterprise LinkedIn account giving a shoutout to @kapa_ai

We’re super AI-native, and our AI tool stack is how we support 100+ customers incl. @OpenAI, @Reddit, and @Docker with just a 10-person team. Here’s what we use: Sales/Go-To-Market: - @MeetGranola ($14/mo): Hands down the best AI note taking app that’s more like AI-improved note taking since it has a handy scratchpad - @NotionHQ AI ($15/mo): To ask “Do we already have a customer that requested feature X” to quickly find the right backlog item without searching through 100s of tickets - @AnthropicAI Claude ($20/mo): Turn meeting transcripts into well-worded follow up email/slack messages based on pre-defined templates (We find Claude is much better for natural writing than OpenAI) Engineering: - OpenAI Deep Research ($200/mo): quickly scan for new toolsets (e.g., best pdf conversion tools) and approaches to problems - OpenAI o1/o3-mini-high ($20/mo): For making code changes and finding bugs (we prefer OpenAI for code, on the other hand) - @Cursor_ai($20/mo): For introducing random bugs and edge cases that make all of our test cases fail. - @kapa_ai (priceless): For “talking” to the dev tools we use like Sentry, Docker and Weaviate with their “Ask AI” widgets on their docs to troubleshoot The common theme here is to find your most manual, pointless process and see how AI can do the heavy lifting. I do not know where we’d be as a company without these tools, tbh.

very biased, but at @kapa_ai we've been helping awesome companies like @prisma @raydistributed @mapbox @CircleCI @FusionAuth @WebinyCMS @trydaily @memgraphdb @OryCorp @algodevs do just this 🫡

@frankdilo @linuz90 @ev is there anything we can do to support @mailbrew coming back to life? 😢

... a rare @bauefius and @emilsnotes spotted out of office👇 link: piped.video/watch?app=deskto…

want to come build advanced RAG pipelines and work on evals with us at @kapa_ai? ✨ we're hiring: workatastartup.com/jobs/6432…

well there's my vote for the coolest thing I'll see all week. 👇 thanks for sharing @MatijaSosic and go check out @WaspLang for you next full-stack deployment.

How have I not seen this - $15B juggernaut @Mondaydotcom mentioned kapa and called us their “Super powerful developer AI” 🔥 Check this out. They’ve built @kapa_ai right into their new API Playground and developer center. I am super proud of this deployment. You can ask it to do anything: - Write API requests - Fix bugs - Explain anything about the API documentation Seeing “powered by kapa.ai” in the corner of the most critically helpful AI assistant for a $15B public company is the type of inspiration I live for.

nothing is more rewarding than seeing @kapa_ai end users saying stuff like this 👇

awesome feedback from a new @prisma developer using kapa on the docs ✨

so freaking cool to continue seeing @kapa_ai mentions in this thread! thanks @swapnilmmane! 🙏

want to come do cool full-stack work with us at @kapa_ai and dip your toes into LLMs in the process? ✨ we're hiring: ycombinator.com/companies/ka…

so @rikschennink officially holds the record from fastest @kapa_ai API implementation🏆 ... 16 mins from API key to prototype🤯

Highly recommended post by @leeerob on making products developers really want.

this has been a fun one to 🚢

Microsoft just rejected a documentation PR because "tables don't translate well to AI chat." This is exactly backwards. Here's why: 1. Tables make technical docs more scannable and easier to understand 2. Modern LLM systems can absolutely parse and understand tables 3. We should build better AI, not worse documentation The simple truth we've learned while building @kapa_ai : What's good for humans is good for AI.

My take: Good docs for humans are good docs for LLMs. We’ve worked with top technical companies like @getsentry, @docker, and @openai to adopt LLMs trained on their documentation. And overwhelmingly, I see that what works well for humans, works well for LLMs. Let me prove it to you. Here’s our official guidance for optimizing technical docs for optimal LLM consumption: - Embrace page structure and hierarchy - Segment documentation by sub-products - Include troubleshooting FAQs - Provide self-contained example code snippets (include imports) - Write text descriptions for images - Define specific acronyms and terms We’ve done this for 100+ teams, and it works. But I read that, and all I can think is “Those are also best practices for writing technical docs, period.” Humans want FAQs. They don’t want to guess what an acronym means. And they want well-defined and structured docs. So…if you’re trying to optimize your docs for an LLM, try to optimize for a human. If you follow the best practices, you’ll create docs that both will love to read.

pretty cool to see @OpenAIDevs community leaders starting to cross-post @kapa_ai answers from the developer Discord to the forum 💜

“Certainly, here is the code snippet!” It doesn’t work ChatGPT “So sorry, you’re right. Try this one” Still doesn’t work “Ah! I see the issue. Try this!” Doesn’t work “Let me try this. Did it work?” No “Ah, found it. This should work” This is the same as the first snippet “My goodness, good catch! What about…

1 year post-@ycombinator and a lot has changed with @kapa_ai as we scale. But this type of direct customer feedback is still the coolest part about being a founder.

super excited to have @kapa_ai help the forward-thinking @AirbyteHQ team scale support to the +10.000 developers in their community 🚀

Companies: “Read the docs, answer is right there” Where the users go instead: - Google - Ask ChatGPT - Stack Overflow - Consult a Ouija board - Random internet forum from 2006 - Install Cursor so it can have a go at it - Train an entire LLM on their codebase - Hope the answer comes to them in a dream - Cold DM the principal engineer on the product - Teach themselves a new programming language where the error message hopefully won’t appear Read the freaking manual.

Exciting news: @redpandadata (just named by Forbes as one of 25 companies most likely to reach unicorn status) is now using @kapa_ai transform their developer experience. 🎉 They're processing hundreds of terabytes of streaming data daily for companies like Midjourney. That means their documentation needs to be incredibly precise and accessible. When we first started working together, they had a clear challenge: "Search results were hard to navigate. It was purely keyword-based search, not context-aware. Users would get a list of results, click one, find it wasn't what they wanted, go back - and lose their whole list of results." Together we integrated an AI assistant across their entire developer platform - docs, support portal, internal Slack, and community channels. Now every technical question gets an instant, accurate response backed by their knowledge base. In 5 months since deploying @kapa_ai: - 5,000+ technical questions answered monthly - 93% of questions answered based on existing knowledge - Nominated for best AI-augmented documentation at 2024 @DevportalAwards But here's what really matters. Technical Documentation Lead Joyce Fee put it as follows: "We started out looking for a better search experience for our users. Very quickly, we realized that this is way more than a search mechanism - users could get specific configuration outputs and code examples in their preferred programming language instantly." Want to see how @redpandadata uses @kapa_ai? Check out: 👉 Their AI-powered docs on the Redpanda docs 👉 Our latest case study (in comments) Huge thanks to the @redpandadata team for trusting @kapa_ai - excited for what's cooking in the coming months.

launch alert: tech support widget powered by gpt-4 we gave @kapa_ai access to all @open_ai: - docs: platform.openai.com/docs - cookbooks: github.com/openai/openai-coo… - forum posts: community.openai.com/ - github issues: github.com/openai demo: kapa-widget.vercel.app/widge… 🕹️

the best kind of @kapa_ai love in the wild. docs AI = faster developer onboarding 🧙

I'm excited to announce that @kapa_ai has been named # 3 on the list of "Top 100 Cloud Challengers" by @NotionCapital and @GoogleCloud! This recognition highlights the most innovative cloud startups across Europe, and we're honored to stand among such incredible companies. While we've built our team in Silicon Valley, our European roots remain a fundamental part of our identity and approach. What makes this achievement even more meaningful is that we've accomplished it with just 14 team members - right at the median size for companies on this list, but delivering enterprise-level impact that typically requires much larger organizations. A huge congratulations to all the other recognized companies! As for us, we'll keep focusing on what we do best: transforming your documentation into reliable, LLM-powered chatbots that deliver real value.

some very nice words from @prisma ceo and co-founder about @kapa_ai🙏

this came out so so well team! 🫡

We love @getsentry at @kapa_ai too 🔥

another @kapa_ai mention in the wild 👀

launch alert: gpt-4 powered docs widget for @discord we gave @kapa_ai access to all discord: - developer docs and api reference 📂 - help articles from @discord_support 🆘 - github issues and bug 🐞 demo: kapa-widget.vercel.app/widge… 🕹️ what framework should we do next?

My AI notetaker caught this feedback, maybe we should update our homepage 🤣

✨ Insanely exciting news: @Logitech just implemented @kapa_ai to supercharge their documentation with Instant Answers. Logitech maintains over 2,000 pages of complex technical docs across their wide product suite - from keyboards to enterprise video solutions. Their users can now get instant, accurate answers to any product question when browsing their business support portal to better self-serve. Fun fact: The entire @kapa_ai team's favorite webcam is the Logitech MX Brio Ultra.

it's been super fun to collaborate with @josephofiowa @braddwyer and team to integrate @kapa_ai into the @roboflow support stack 🚀

grounded code gen + up-to-date knowledge sources ✨

The best demo I've ever seen was in @pablorpalafox and @palafoxjavi's kitchen last year when they showed me their initial voice agent prototype they had been hacking on. So freaking cool to see how quickly they've gone from that initial prototype to automating more than 1M calls for freight companies in the last year. The @kapa_ai team will still beat you all in tennis again though. :)

1/ we gave gpt-4 access to all the latest @nextjs and @vercel docs, blogs and templates and created a custom @kapa_ai discord bot 🧵 comparison thread versus off-the-shelf gpt-4 model below

had so much fun recording this with the llama man himself @jerryjliu0 🦙🦙

“I don’t know” is a completely fine answer for an AI agent to give, btw. I’d rather my agent answer that than do something like: - Make something up - Try to do something it shouldn’t do - Accidentally reveal sensitive information We take a lot of care around our prompting to make sure the AI agent understands exactly where its abilities stand. Sure, it’s extra work, but I’ve seen too many horror stories to go back at this point. Your employees can say “I don’t know”...let’s extend the same graces to our AI agents.

calling all mega talented engineers out there. insanely cool opportunity.👇🏻

Search is AI’s killer use case. Glad to see some coverage from Sequoia Capital on the same industry bet we’re making. “AI search is an informational technology based on LLMs that can read and semantically understand knowledge. For white collar workers, this will be a huge boon.“ People have been using ChatGPT like a search engine since it launched. It’s no surprise why: - Plain English answers - Whatever technical depth you need - Drawing on deep knowledge from across subject matters AI isn’t going to change the simple fact that people want information. They’ll just have higher expectations for accessing that information. And if you have proprietary information (like your docs), you’ll want to make sure AI can surface it correctly. What AI search tools do you use most?

docs are hard.

getting roasted by @mwseibel in weekly group office hours was both humbling and honestly one of the best things that ever happened for @kapa_ai. thank you.

💜

ouch. i don't think I've ever seen @AnthropicAI say literally everything is down before.

Community is such an underrated part of the journey. I will always firmly believe YC is the best way to increase your chances as a founder and as a startup. I know the deadline for the Spring batch just passed, but if you’re at all interested in being an entrepreneur…apply for the summer one. You won’t regret it.

btw if you're looking to turn your product knowledge base into an AI onboarding/support assistant head over to kapa.ai/request-demo and we'll set one up for you (like we have @Docker and @mondaydotcom)

we made it! thanks @barnji for the screengrab. 📸

ANNOUNCEMENT: @kapa_ai is hitting the road, at KubeCon in London, April 3-4. We’ll be talking AI docs with a bunch of cool customers including @CircleCI, @GetSentry, @Grafana, @RedPandaData, @Sonatype and…possibly you?? We’ll bring a team of the smartest kapa folks and we’re super excited to meet the community, and show everyone how powerful your docs can become with an “Ask AI” assistant. If you want to catch up and/or talk all things turning technical knowledge bases into AI assistants we'd love to get you coffee. Feel free to DM me! And if you haven’t heard of AI-powered assistants for your docs, well…you won’t need that coffee, the excitement alone will power your day. 😉 See you at #KubeConLondon!