Flowpoint

Flowpoint is best understood as a deployment decision that happens to include an AI agent, not an AI analytics tool that happens to offer on-prem. The homepage is unambiguous about which one leads: three steps, and step one is "License Flowpoint and deploy it inside your own environment — your servers, your cloud, your network." The agent arrives afterwards, pointed at a local model running in your network or a cloud model like Claude. What that buys you is a clean privacy story. One tracking script goes before the closing </head> tag; everything after that — collection, storage, processing — stays inside your walls. No shared tenancy, no third-party processor, and Flowpoint states plainly that nothing is sent to them or any external service. For a team whose data-residency review is the actual blocker, that answer is short and checkable, which is worth more than a feature comparison. The MCP server is the part most buyers will underestimate. Flowpoint exposes your traffic, events, funnels, and dashboards as MCP tools, so Claude, Claude Code, Cursor, or any MCP-compatible client can pull from your own data and build dashboards in natural language. If your team already works inside those tools, analytics becomes one more thing the agent can reach rather than a separate console to log into. Note what the vendor says about inference: the agent builds the dashboard interfaces, and you decide whether the model runs locally or in the cloud. Day-to-day capability is solid rather than sprawling. Real-time visitors and sessions, conversions and custom events through the global flowpoint() function, filters by device, browser, country and UTM source, conversion funnels for finding drop-off, and PDF export for the people who still want a document. The marketing pitch includes a Flow Builder for agentic process automation, Page Inspector scans for SEO and accessibility issues, plus the ability to connect Google Analytics or upload a CSV and question it in chat. Install guides cover HTML, React, Next.js, WordPress, Shopify, Framer, and Angular — the docs walk you through a sub-two-minute script paste. Where it asks something of you: this is licensed software, not a signup. The pricing page takes a request, an intro call follows, and then you get a scoped build tied to your deployment and model choice. That is a real procurement cycle, and you or someone on your team has to run and maintain the install afterwards. There is also no broad connector marketplace here — the documented ties are Claude, Claude Code, Cursor, and Google Analytics, with the MCP server as the general-purpose route for everything else. And because the license is scoped per deployment, what you can run is whatever your agreement says, not a public tier list.