Privacy-first analyticsSay hello to your little friend.

The Google Analytics replacement that never snoops. A tiny script that tells you who or what arrived, where they came from, what they did, and what worked.

Private beta. Accounts are by invite for now, and we'll email you when there's a spot. See the install.

  • 4.0 KB.The whole script, gzipped.
  • No cookies by default.Aggregate mode keeps no visitor or session key, anywhere.
  • Bots in the open.Crawlers and AI agents get their own report, and a door if you want one.

All the answers, none of the setup.

No dashboard to build. Add the script and it's ready.

  1. Who or what is arriving?

    People, search crawlers, AI crawlers, AI agents and uptime monitors, each counted in its own lane.

  2. Where did they come from?

    Referrers, campaigns and channels. A hidden source shows as "Direct or unknown".

  3. What happened after they arrived?

    The pages they read, the actions you named, and, with consent, a readable timeline of each visit.

  4. Where did it stall or succeed?

    Goals, funnels, and outcomes your server confirms.

  5. How much can you trust this?

    Every number carries its definition, freshness and caveats. Missing data shows as unavailable, never as zero.

Small enough to forget it's there.

Every visitor downloads your analytics script. Ours has a 4 KB gzipped budget: Plausible and Fathom territory, and under a fortieth of Google Analytics on the sites we measured.

Gzipped size of each analytics script
ToolSize, gzip
Google Analytics 4gtag.js for a real GA4 property163 to 181 KB
PostHogarray.js, the file the snippet loads95.9 KB
Little Friendlf.js (Full), measured at build4.00 KB
Fathomscript.js2.0 KB
Little Friend Litelf-lite.js, pageviews and custom events1,895 bytes
Plausiblescript.js, the default script1.2 KB
Gzip sizes, measured September 25, 2026, with 1 KB as 1,024 bytes. Google Analytics shows the range across eight real sites. Little Friend's comes from the built lf.js. Lite is 3,178 bytes as written, 1,895 bytes gzip and 1,701 bytes brotli. How we measured

Every visit, step by step.

See what each visitor did, in order. Name the moments that matter, like choosing a plan or starting a signup, and every visit reads as a short timeline.

  • Steps are names you declare, like billing-period.
  • Journey mode is opt-in, and it can wait for consent.
One visit, as Little Friend sees itSample
  1. Opened /pricingSource: newsletter
  2. Chose the annual planButton: billing-period
  3. Started signupForm: signup
  4. Signup failedReason: validation
  5. Signup confirmedConfirmed by your server

A journey never stores what the visitor typed, what the page said or where the mouse went.

See which AI is reading your site.

Google Analytics quietly drops known bots. Little Friend shows which crawlers and AI agents read which pages, and how we know who they are. Google's note on bot exclusion

Agents, last 7 daysSample

12,480 browser page views

9,791 automated requests

Counted separately from page views.

VisitorKindHow we knowRequestsMost read
GooglebotGoogleSearch crawlerVerified network4,210/docs
GPTBotOpenAIAI crawlerVerified network2,388/docs/install
ClaudeBotAnthropicAI crawlerSelf-declared1,406/blog
ChatGPT-UserOpenAIUser-directed AI agentVerified network312/pricing
Uptime checkYour monitorUptime and monitoringSelf-declared288/health
UnknownNo claim madeSuspected automationNo evidence1,187/login
  • Every label shows its evidence. A user agent is a claim. A matching DNS record or published IP range is a verified network. A valid request signature is stronger still.
  • AI citations aren't tracked. GPTBot reading your docs doesn't mean an AI answer used them.
  • Most crawlers skip JavaScript. To see them, add a Vercel log drain or send requests from your edge.

AI agent analytics Agents docs

Know your agents.

Let the agents that bring customers in. Slow down the ones that only take. Little Friend checks each agent's signature, names who runs it, and tells you when it acts for a signed-in user. It's Know Your Agent (KYA) for your own site, next to the numbers you already read.

The passport

Who the agent is, who runs it, what it says it's for, whether that checks out, and who it acts for.

ChatGPT agentOpenAISample
Run by
OpenAI
Says it's for
AI agent or agentic browser
Checks out
Verified signatureSigned with a key OpenAI publishes
Acts for
A signed-in userVisa Trusted Agent Protocol tag
robots.txt
Stayed out of every disallowed page
At your door
Let in

The door

Rules for who gets in, how often, and who has to sign. See what they would have done to last week's traffic, and run them in dry run until you trust them.

Your rulesDry runSample
  1. Block AI training crawlersBlock
  2. Block agents that fail their operator checkBlock
  3. Checkout and accounts need an agent acting for a signed-in userRequire a signature
  4. Slow down agents that are not verifiedSlow down, 60 a minute
  5. Everything elseAllow

Last 7 days in dry run: 3,794 requests would have been blocked and 212 slowed down.

The ledger

What each agent takes and what it brings back: visits from its assistant, the goals those visits reach, and the revenue your server confirmed.

Last 30 daysSample
OperatorRequestsVisits sentPurchasesRevenue
OpenAIGPTBot, ChatGPT agent11,6402129$1,340
ByteDanceBytespider8,905No assistant sends visitors back
AnthropicClaudeBot, Claude-User6,120382$180
PerplexityPerplexityBot, Perplexity-User1,310964$520

Revenue counts purchases your server confirmed, in journey mode.

  • The door is optional. Leave it out and Little Friend just counts. Add it with a few lines in your Worker, proxy or server.
  • People always get in. The door only answers automated clients. Every browser is let through.
  • Who they act for stays with you. The door sends us its decision, and nothing about the person behind the agent.

Set up the door How agents prove who they are

Private by default. Detailed by consent.

Aggregate is the default. Journeys are there when your visitors say yes.

Aggregate Default

Counts, with no way to follow anyone.

  • Views, sources, pages, devices, countries and events
  • No cookies, no visitor key, no session key, anywhere
  • Source to goal, within a single page view
  • No timelines, sessions or returning-visitor counts

Journey Opt-in

Readable visits, only with consent.

  • A random session id per browser tab
  • Ends after 30 minutes idle, never lasts past 24 hours
  • Timelines, funnels and per-visit sources
  • Raw events kept 7 days by default or up to a year, totals 13 months
  • Optional session replay, with every word masked until you show it

Count the sales your server confirms.

Send the outcome from your server with a secret key, and Little Friend keeps it apart from browser clicks. Mark a goal as server-confirmed and clicks stop counting toward it.

38clicks on Buy, seen in the browser

31purchases your server confirmed

Sample
Send a confirmed outcome
// On your server, after the payment really went through
await fetch('https://in.littlefriend.io/v1/server', {
  method: 'POST',
  headers: {
    authorization: `Bearer ${process.env.LF_SERVER_KEY}`,
    'content-type': 'application/json',
  },
  body: JSON.stringify({
    v: 1,
    events: [{
      id: order.id,
      name: 'purchase',
      correlationId: order.checkoutId,
      value: { amount: 4900, currency: 'USD' },
    }],
  }),
});

Any language that can send an HTTPS request can use it. Goals and server events

Install in one line.

Paste it, then watch it work. Page views, route changes in single-page apps, outbound links and downloads are counted without more code.

In your site's <head>
<script defer src="https://cdn.littlefriend.io/lf.js" data-site="lf_YOUR_SITE_KEY"></script>
  1. Paste the tag with your site key.
  2. Open the live install panel in your dashboard.
  3. Visit your site. Your first page view shows up, with every field we kept and any we dropped.

Install guide Custom events and data-lf

How it compares.

Every claim about another tool links to its source.

FeatureLittle FriendGoogle Analytics 4PostHogFathomPlausible
Script size, gzipUnder 4 KB (our budget)163 to 181 KB, measured on 8 real sites95.9 KB for the first file2.0 KB1.2 KB
Bots and AI crawlersAn Agents report with evidence for each one, and an optional door at your edge that lets them in, slows them down or blocks them. Crawlers are counted from a Vercel log drain or your edge.Known bots removed automatically. You can't see how many.Known bots blocked in the browser SDK. Forwarded server logs can be classified by user agent and published IP ranges (labeled a new feature).Filtered outFiltered out, including data center traffic
CookiesNone in aggregate mode, the defaultYes. Consent mode can send cookieless pings when storage is denied.First-party cookie by default. A cookieless mode is available.NoneNone
Available todayPrivate beta, by inviteYesYesYesYes

Little Friend vs Google Analytics Little Friend vs Plausible Little Friend vs Fathom

Questions.

Something missing? Email hello@littlefriend.io.

Can I use Little Friend today?

Little Friend is in private beta, and accounts are by invite for now. Join the waitlist and we'll email you when there's a spot.

Do I need a cookie banner?

Little Friend sets no cookies in aggregate mode, the default, and keeps no visitor or session key anywhere.

Whether you need a consent notice depends on where you operate and what else your site does, so we can't promise you won't. We list what we collect so you and your counsel can decide. Journey mode is opt-in, and it can wait for consent.

Do you record sessions?

Only if you turn on session replay. It is off by default, works only in journey mode, and waits for the same consent. A recording shows each page's layout with every word masked until you choose which text to show. Form values, images, video and iframes never leave the browser.

A journey on its own is a short list of named steps, like "Opened /pricing" and "Signup confirmed", with no page text, keystrokes or form values.

How is this different from Plausible or Fathom?

They're great: small and cookieless. Little Friend is in the same size range and adds four things: journeys, an Agents report that shows crawlers and AI agents with evidence and lets you choose which get in, goals your server confirms, and an MCP server so AI assistants can read your numbers.

Can I see AI crawlers like GPTBot and ClaudeBot?

Yes, with a Vercel log drain or by sending requests from your CDN or server. Most crawlers never run JavaScript, so a browser script alone can't see them.

Can I block AI crawlers or slow agents down?

Yes, with the door in @littlefriend/edge. It runs in your Worker, proxy or server, and your rules decide which agents get in, how often, and which have to sign their requests.

It starts in dry run, so you see what each rule would do before it changes anything. People in browsers always get in.

Does it work with single-page apps?

Yes. The script counts History API navigations as page views, so most routers need no extra code. Other routers report a view with one call.

Declare route templates like /docs/:slug, and ids in URLs never turn into thousands of separate pages.

Will Claude or other AI agents be able to read my analytics?

Yes. Little Friend has a read-only MCP server. Connect Claude, Cursor or any MCP client, ask "what changed this week?", and get the same numbers and definitions as the dashboard.