Mixpanel · Cloudflare
Mixpanel ships "the product intelligence system for the AI era" — Product Analytics, Web Analytics, Session Replay, Experiments, Metric Trees — plus three new AI surfaces: Mixpanel AI, Agent, and Headless. Cloudflare's developer platform is the runtime that keeps the inference economics, edge query latency, and the Rolldown-built front-end on one consistent stack.

Product intelligence for the AI era.
The runtime underneath it.

The pitch on mixpanel.com is direct: always-on product intelligence that proactively surfaces insights, diagnoses problems, and recommends what to do next. Mixpanel AI does that across billions of events per month — and Mixpanel Agent + Headless extend that intelligence into autonomous workflows. Each new surface is multiple model calls per query, multiplied across thousands of customer workspaces.

Billions
of events per month, sub-second queries
3 AI surfaces
Mixpanel AI · Agent · Headless
12,000+
in the largest analytics community

Three things converged in the last week that change what's possible underneath the Mixpanel AI stack — and one of them is about the front-end build pipeline you may not have realized is now in our stable.

GA · June 2026 Your AI bill is out of control. Cloudflare can fix it now.
AI Gateway ships dollar-denominated spend limits per customer workspace, identity-driven budgets, PII redaction, and automatic fallback routing — across every model provider, in one log line. For Mixpanel AI / Agent / Headless specifically, this is the runtime piece that makes "always-on intelligence across billions of events" economically defensible at the per-workspace level. Read the announcement.

Three primitives that map directly to how Mixpanel actually delivers Mixpanel AI:

AI Gateway + Workers AI — under every Mixpanel AI / Agent / Headless inference call: per-workspace cost attribution, PII redaction before the model sees customer event data, fallback routing across providers. The answer to "what's the per-customer AI cost of always-on intelligence?"
Workers + Workers Analytics Engine — for the event-ingestion edge layer: capture events at the closest of 330+ POPs, pre-aggregate at the edge, only ship rolled-up signal to the central query engine. The pattern that keeps sub-second queries sub-second as event volume grows.
VoidZero (Vite, Rolldown, Oxc) — the team behind your Framer + Rolldown front-end joined Cloudflare on June 4. The bundler running mixpanel.com right now is built by people on the same team that fronts your domain.

One question for the Mixpanel platform team:

Is the bigger near-term pain on the AI-economics side — keeping per-workspace inference cost predictable as Mixpanel AI scales to every customer — or on the edge-latency side — pushing event ingestion and pre-aggregation closer to the user than the central query engine allows? 20 minutes to find the right starting point.

Deeper Dive

The full architecture, ready when you are

A detailed primitive-by-primitive mapping is in progress — including the Mixpanel AI × Cloudflare primitives table, the request-flow diagram for an Agent inference call on Cloudflare, the AI Gateway cache math for billions of events, the Rolldown / Vite roadmap angle, and the 90-day rollout. Tell me which slice would be most useful first.

Request the expanded version →
Grab 20 minutes →

One observation worth flagging: mixpanel.com is already fronted by Cloudflare today. So this isn't an "add a vendor" conversation. It's a "turn on what's already in the stack — and put it under Mixpanel AI too" conversation.

Matt Holscher Cloudflare · Developer Platform