See what moved.
Ask why.

AI-first web analytics: dashboards when you want to look, an AI agent when you want to know why.
No cookies, no personal data.

Works in Claude, ChatGPT, Gemini and other AI apps

fiveb.ar/example.com

example.com 7 on the site now Last 30 days

goalSignup×deviceMobile×

Conversions: 94; ▼ 41%
Events: 101; ▼ 40%
Conversion rate: 1.7%; ▼ 1.2 pts

0 2 4 6

Since 9 Sept

29 Aug 5 Sept 12 Sept 19 Sept Today

Connected to fivebar and GitHub

Signups on phones are down 41%. Why?

fivebarexplain_change
fivebarget_site_speed
GitHubpull requests

/pricing got slow on 9 Sept, the day #412 shipped the new calculator. Taps there take 520 ms now, up from 180, and that's about 55 of the 66 missing signups.

Connect it once

Add fivebar to Claude, ChatGPT, Cursor or any app that supports MCP. No API keys, no CSV exports, no pasting numbers into a chat.

Add the address

https://fiveb.ar/api/mcp
Sign in and choose Allow
Start asking

Step by step in

The questions a dashboard can't answer

Your AI reads fivebar next to your code and your team's chat, and puts it together

What's behind this week's jump?

fivebarexplain_changelast 7 days
fivebarget_site_engagementsource Substack
fivebarget_site_summarysource Substack, goal Signup

A Substack post linked your pricing guide on Tuesday: 3,100 visits, most of them from the US. They stayed, too: over three minutes on the guide, and 41 signed up. Worth a thank-you to the writer, and a follow-up post.

Find what's working

It traces a rise to where it came from, and checks the people it brought stayed.

Which guides turn readers into signups? Make the rest more like them.

fivebarget_site_breakdownentry pages, goal Signup
fivebarget_site_breakdownscroll depth, pages in /guides/
Editcontent/guides/3 files

The ones with a worked example: they convert at 4.1%, against 1.2% for the rest, and are read to the end twice as often. I've drafted a worked example for the three without one, ready for you to review.

Do more of what works

In your editor, it finds what your best pages share, and brings the rest up to it.

Do people from Hacker News pick a different plan? Should we post there more?

fivebarget_site_breakdownplan, goal Signup, source Hacker News
fivebarget_site_breakdownplan, goal Signup, source Google
fivebarget_site_summarysource Hacker News, last 90 days

example.com90 days up to yesterday · 1 Jul – 28 Sept 2026

sourceHacker News

Pageviews: 5,852; ▲ 138%
Visits: 4,370; ▲ 131%
Visitors: 4,212; ▲ 130%
Bounce rate: 61%; ▲ 4 pts
Visit duration: 1m 5s; ▼ 12%

0 1,000 2,000 3,000

6 Jul 27 Jul 17 Aug 7 Sept 28 Sept

Open in fivebar

Hacker Newsstories linking example.com since 1 July

Very: 41% of their signups choose Pro, against 12% from Google. But don't count on the next post. Each of your three brought fewer visits than the last, 2,100, then 1,300, then 700, and only the first, your Show HN, reached the front page.

The dashboard, in your chat

Claude, ChatGPT and VS Code draw a site's summary with fivebar's own tiles and chart, beside the answer.

Installed by your agent. Checked by fivebar

Paste one line into Claude Code, Cursor or Codex. It finds the layout every page shares, adds the tag, asks fivebar whether it's counting, and sorts out your Content Security Policy if it isn't.

Paste into your coding agent

Add fivebar analytics to this site, following https://fiveb.ar/docs/coding-assistants.md

fivebarget_site_setupexample.com
Editsrc/layouts/Base.astro+1
fivebarcheck_installexample.comBlocked by CSP
Editpublic/_headersfiveb.ar in script-src and connect-src
fivebarcheck_installexample.comWorking

Done. Your Content Security Policy was blocking the script, so I've added fiveb.ar to it. fivebar says it's counting.

Rather do it yourself? It's one script tag.

A dashboard worth opening

Everything your AI reads, drawn for people, beside how it looked last time

fiveb.ar/example.com

example.com 7 on the site now Last 30 days

Pageviews: 30,687; ▲ 12%
Visits: 12,577; ▲ 9%
Visitors: 9,600; ▲ 11%
Bounce rate: 41%; ▼ 3 pts
Visit duration: 2m 34s; ▲ 6%

0% 25% 75% 100% 3s 10s 30s 3m 10m Avg 52s Avg 48% SkimmedRead closely Left earlyStayed near the top Time on pageScroll depth

/ /blog/cookieless-by-design3m 10s on screen, 76% of the way down /docs/getting-started /blog/ /pricing /docs/api /blog/tally-marks /about /changelog /docs/privacy /contact /terms

Goals

Name what counts, and see which visits got there.

Filters

Click anything. Then something else.

LCP: 1.9 s; ▼ 0.3 s; Good
INP: 240 ms; ▲ 40 ms; Needs improvement
CLS: 0.03; No change; Good

75th percentile of page loads

Slow pages

Timed in your visitors' own browsers, not a lab.

Your AI sees trends, never people

No cookies, and nothing left on anyone's device. Each visitor is a one-way hash, salted with a secret that's thrown away at midnight, so there's nothing personal for fivebar to keep or for an app to read.

Apps you connect can read your sites' figures and nothing else, and you can disconnect them from Settings at any time.

Come back tomorrow and you count again. A fair price.

IP addressBrowserToday's salt

One-way hasha41e 9f2c 07d1 5b3e

Counted once today

Midnight

~~Today's salt~~ overwritten. Yesterday's hashes can never be made again.

Room for a few more

fivebar is in private beta. Leave your email and we'll be in touch when it's ready.