Frederik

Introducing Agent Analytics for @kapa_ai: you can now view how Claude Code, Codex, and Cursor interact with your product. Here's how: 🧵 Quick context: 100+ companies (n8n, ClickHouse, and more) power their docs MCP servers with kapa. These integrations are growing FAST. When an AI agent has a question about your product, it queries your curated KB via MCP directly to get grounded answers instead of stale training data or flakey web search. You could always see these queries. But now the analytics got a serious upgrade: - Query volume broken down per MCP client (Claude Code vs Cursor vs Codex...) - Dedicated agents view to filter and read agent conversations - Export interactions to work out how agents actually use your product Agents are becoming some of your heaviest docs users. Now you can see exactly what they're up to. Try it: kapa dashboard → Agent Analytics

We just hit the front page of Hacker News again - this time with how we taught a small LLM to throw away 68% of our RAG context. Here's how we implemented pruning:

Today we're launching Kapa for Agents: all your product knowledge, in one tool call. As models get smarter ... and pricier ... (looking at you Claude Fable), context is becoming the bottleneck. Here's why: Say you've built an agent in your SaaS app with some tools. Then a user asks "How do I enable SSO?" but you haven't given the agent a tool to fix that. It doesn't matter how smart your agent is, it hits a dead end. Instead, with Kapa, you can add easily add a single knowledge search tool so your agent can read your docs, code, tickets. Real-world agents use this in +40% of interactions to improve planning and avoid dead ends. TL;DR: we spent 3 years building the best agentic retrieval platform that: → Finds the right source ~2x more often than web search or a DIY RAG pipeline → Tells you what your agent couldn't answer, and exactly how to close the gap → Connects 30+ sources in one click, synced in real time so knowledge doesn't go stale → Works with any agent: product copilot, support agent, or Claude Code Teams like @tweetsbyport, @airbytehq, @circleci and @matillion are already using Kapa's Retrieval MCP and API to build in-product copilots, support agents, RFP tools and coding assistants. ... and today, its available to everyone at kapa [.] ai / agents.

TanStack is one of the fastest-growing JS frameworks in the world. 375M+ weekly downloads. Used by teams like Anthropic. Tons of new launches like TanStack AI SDK And now their docs have a grounded AI agent powered by @kapa_ai. Why? Because @tan_stack ships. The docs change constantly. The product evolves constantly. And if developers - or AI coding agents - are not reading the latest docs, they get left behind. That leads to outdated answers, poor implementations, and a lot of wasted time. So the TanStack team needed an AI assistant that could: → Index thousands of docs pages in real time → Give answers grounded in official sources → Cite where the answer came from → Say “I don’t know” when the docs don’t support an answer → Help the team spot gaps in their documentation through analytics That’s now live. Huge credit to Anton Malling from kapa team and @tannerlinsley for the awesome deployment. We're just getting started!