Klavis AI (YC X25) reposted this
Excited to share that Klavis AI (YC X25) has successfully completed its SOC 2 Type 2 audit! We’re committed to upholding the highest standards of security, availability, and confidentiality for our customers.
Klavis AI (YC X25)
Software Development
Open source MCP integration platform that let AI agents use any tools reliably, at any scale!
About us
Klavis AI provides open source MCP integration for AI applications. Integrate enterprise-grade MCP servers with your AI application in a minute, and scale to millions of users. Available in UI, API and open source!
- Website
-
https://www.klavis.ai/
External link for Klavis AI (YC X25)
- Industry
- Software Development
- Company size
- 2-10 employees
- Type
- Self-Employed
- Founded
- 2025
Employees at Klavis AI (YC X25)
-
Lawrence Lou
Very active VC Investor | Early-Stage AI Advisor | AI builder & strategist | AI Product Leader | ex-Airbnb, ex-Microsoft
-
Xiangkai Zeng
MCP integration platforms that let AI agents use tools reliably at any scale. | Co-Founder @Klavis AI (YC X25), ex-Google DeepMind
-
Zheng Liu
MCP layer for AI agents to master tools at any complexity | Founding Engineer @Klavis AI (YC X25)
-
Rohan Sharma
Problem Solver | OSC- yc(F24)UseAutumn, yc(X25) Klavis Ai | Ex- Backend Intern @GoQuant | Ex- Fullstack Intern @getLiquefy | Newbie Codeforces |…
Updates
-
Klavis AI (YC X25) reposted this
The latest Andrej Karpathy podcast hit different... One line stuck with me: "Progress from a working demo to a reliable product is a 'march of nines.'" - getting something to work 90% of the time is just the first 'nine.' - each subsequent nine (99%, 99.9%, 99.99%) requires massive, constant effort. You've seen the AI agent demos everywhere, fancy MCP integration, tool calling, perfect executions. Then production hits. Hallucinations. Infinite loops. Different outputs for the same input. Tool accuracy at 20%. Monday it works. Tuesday it's dead. The gap between demo and production? It's not a final sprint. It's where most teams realize they're only 5% done. LLMs are brilliant improvisers but terrible employees. They don't fail because they can't do something, they fail because they can't do it consistently. Each 'nine' isn't just more testing, it's architectural redesign, observability systems, graceful degradation by design. Demos are dangerously easy, production agents are brutally hard. Again, we're in the decade of agents, not the year of agents.
-
-
NEW: now your AI agents can handle complete Microsoft 365 operations through Klavis AI (YC X25), use it out-of-the-box without worrying about authentication and context window tool limits! 1/ automatically manage files and folders across OneDrive 2/ generate AI-powered emails, drafts, and organize Outlook workflows 3/ sync team communications and coordinate channels in Microsoft Teams 4/ access comprehensive Microsoft 365 libraries and searchable content real-world examples: "share client presentations from OneDrive, send them via Outlook with personalized messages, and notify the sales team in Teams about deal progress" Enjoy!
-
-
Fireflies.ai and Fathom.ai have been integrated into Klavis AI (YC X25)! Now your AI agents can handle meeting note operations through Klavis, use it out-of-the-box without worrying about authentication and context window tool limits! 1/ Automatically transcribe and summarize meetings 2/ Generate AI-powered action items and highlights instantly 3/ Sync meeting insights to CRMs, project management tools, and workflows 4/ Access comprehensive meeting libraries and searchable transcripts Use Klavis AI in your AI agent product to connect it to meeting notetakers alongside other tools like Slack, Jira, and Notion to build powerful productivity assistants today!
-
-
so build your own CRM with Klavis in 1 day ?!
Confession: I am terrible at following up. Conversations with users and customers would get buried in my inbox, and I'd let important opportunities fall through the cracks. The old "process" was a nightmare. I was constantly jumping between cold emails, slack, LinkedIn, etc.., trying to remember who I needed to ping next. It was manual, stressful, and inefficient. I needed a central brain to manage it all. So we built a 4-step system: 1> SYNC: First, it connects to all our channels (Gmail, Slack, etc) and syncs the conversations into a Notion workspace. Everything in one place. 2> IDENTIFY: Next, the system automatically flags any lead that hasn't been contacted recently. It knows who's "going cold." 3> GENERATE: This is the magic. LLM reads the entire conversation history and writes a personalized, context-aware follow-up draft. 4> REVIEW: The draft gets sent to me in Slack for a quick, 1-click review and send. This keeps the final human touch, without the manual effort. How does it connect to everything so easily?? The secret is the MCP. Instead of wrestling with 3 different APIs, our app makes one simple call to our MCP server. The results have been a game-changer: - we never miss a follow-up. - we're saving hours of mental energy every week. - we have a scalable system I can easily add more integrations to.
-
-
Klavis AI (YC X25) reposted this
Many founders make the mistake of assuming the sales process is over when they sign a contract. Think of "signed contract" as the midpoint of your funnel, not the end. At the bottom of the funnel, you should always: 1> Lock in timelines upfront: Before you celebrate any deal, map out your customer's integration roadmap and identify exactly which resources they need to assign. 2> Turn support into your oncall: Whether it's Slack, Teams, or email - their notification becomes your pagerduty alert. No tickets sitting overnight. 3> Define "done" before you start: Understand what a "completed" implementation actually looks like. Get specific, get measurable, then move fast toward your customer's goal. 4> Treat feedback like gold: Every support interaction teaches us something. We use those learnings to make the product better for all our users. At Klavis AI (YC X25), we firmly believe customer support is the most undervalued moat for startups, and we even forward deploy after closing the deal. The job isn't done until your customer has fallen in love with your product and is getting value through habitual usage.
-
-
Today, OpenAI is launching AgentKit, a complete set of tools for developers and enterprises to build, deploy, and optimize agents - https://lnkd.in/gxTnr35r Building it with Klavis AI (YC X25) Strata now, no context overload with tools at any scale!
-
-
4,300 requests handled in .03 seconds This is the performance of Klavis AI (YC X25) Eragon has partnered with Klavis to improve reliability and performance across all MCP connections. You can now query in one workspace: "Tell me about customer usage across opportunities we are negotiating with and break it down by region. When you are done send a report to the executive team." Special thanks to Zihao Lin and Xiangkai Zeng and to the Eragon team for relentlessly shipping to our customers Zhiyang Zhang, Andrew Yang, Qilin Wu
-
-
Competition is for losers Yesterday we met with ACI.dev founder William Lee. What I expected to be a bit awkward turned into one of the most valuable conversations I've had. We exchanged real founder stories, discussed market challenges, explored our customers’ pain points, and shared MCP insights, from developer use cases to enterprise governance. Here's what I learned: we're both fighting to solve the same fundamental problem - making AI agents more reliable and secure when using tools. The real competition isn't each other. It's the status quo. Instead of fighting over market share, we both focus on expanding the whole tool use ecosystem. and make MCP adoption faster and easier for everyone. Sometimes the best way to win isn't to beat your competitor. It's to make the whole space bigger together
-