Describe the data you want in plain English and BrowserAct builds a reusable scraper that runs in a real browser, with proxies and CAPTCHA handled.
Summary
Warehouse-native product analytics that joins AI agent traces to user behaviour, so you can see which tool calls actually convert.
Screenshots
Description
Kubit is a product analytics platform built for teams shipping AI agents, and its premise is that observability and analytics have been answering two halves of the same question. An APM dashboard tells you a tool call took 4.2 seconds; it does not tell you that the user re-prompted twice and then abandoned checkout. Kubit joins the agent's OpenTelemetry trace to the same user's downstream behaviour so both live in one funnel.
How it works
Kubit is warehouse-native and zero-copy: it queries your existing data warehouse in place rather than ingesting a copy, so personally identifiable data never leaves your own infrastructure. Agent traces arrive over standard OpenTelemetry rather than a proprietary SDK, and the platform also bridges CDPs and other event sources, which means instrumentation you already have generally counts.
What you can actually ask it
- User–agent funnels — where a hallucination or a bad tool call breaks a conversion path, step by step.
- Cohorts spanning both worlds — segment on agent behaviour and product behaviour in the same definition.
- Cost and latency against outcomes — tie p95 latency or token spend to DAU, retention and revenue instead of looking at them in isolation.
- Intent and sentiment enrichment on traces, which Kubit positions as an alternative to running expensive LLM evals for the same signal.
- Re-prompt and rage-click detection, the friction signals that infrastructure monitoring is structurally blind to.
Headless mode and coding agents
Kubit can run headless and expose its analytics to coding agents over MCP, so Claude Code and similar tools can pull a funnel, act on what it shows, and then use the same metric to verify the fix landed. The published screenshots show exactly this loop — a kubit-report skill invoked from a terminal agent, returning a channel-level conversion table.
Pricing
Starter is free and pay-as-you-go at $0.008 per monthly tracked user with 10,000 MTU included, unlimited seats and unlimited events. Pro is $199/month plus $0.006 per MTU with 100,000 MTU included and bring-your-own-warehouse. Enterprise is custom, aimed at 1M+ MTU with a dedicated solution architect. Every tier includes the AI features, headless mode and the OTel/CDP integrations, and there is a 30-day unlimited trial with no card required.
Kubit lists Miro, Samsung, Paramount, Skyscanner, CBS, Pluto TV, Synchrony and Serko among its customers, which puts it firmly in the enterprise self-service analytics bracket rather than the startup-dashboard one.
Reviews
Similar App Suggestions
Session replay, error tracking and vision-AI agents in one SDK — agents watch real user sessions on a schedule, file the bugs they find and post to Slack.
Build systematic investment strategies with AI, validate them on out-of-sample and live paper data, then let an MCP-connected agent mirror them in your own brokerage account.
Privacy-first search infrastructure for AI agents, with smart routing, structured results, and native API, MCP, and Skill access.
One REST API that scrapes, crawls, and extracts the web into clean Markdown and schema-validated JSON for AI agents.