cheeseAnalytics, customer data and growth campaigns in one workspace your AI agents can run.
cheese records every visit and product event, turns them into people, sessions and segments, explains what changed on your site, and generates the next campaign from the evidence. Talk to it from Claude, ChatGPT or any MCP client. Your data stays one file you own.
- No card, live in two minutes
- EU-hosted, no raw IP addresses stored
- Bot and AI-crawler traffic kept out of your numbers
- GDPR erasure and export built in
What happens inside the cheese.
Follow one stream of visits from the snippet on your site to a campaign you can measure. Every step is a screen in the workspace and a tool your agent can call.
Every visit, every event.
One snippet sends page views, clicks and the product events you name, batched and consent-aware. Bots and AI crawlers are classified at ingest and never reach a human report.
- page view
- click
- signup
- bot, filtered out
Events become people.
Anonymous visitors get a profile from their first page view. When they sign up, their history joins at query time, so nothing is rewritten and every merge can be undone.
- identified
- anonymous
People become audiences.
Attribute rules, behaviour (did, did not, how often, A then B) and groups compile to one SQL pass and refresh every minute. Preview who is in before anything is sent.
- visited /pricing, no signup
- activated
- at risk
Audiences become campaigns.
Email, webhooks, ads and on-site layers, drafted from the evidence and steered by your brand guidelines. A holdout stays home, so every result is measured rather than assumed.
Every question a growth team asks, answered from the same events.
Four connected layers. Each one reads the data the layer below produced, so a campaign is always traceable back to the visits that justified it.
What changed on your site, and why.
Google Search Console joined with your own landing sessions and one chosen goal, per page. Every number is labelled with what it measures and the period it covers.
- What changed · click changes split into an impression term and a CTR term, page by page.
- Opportunities worklist · explicit screening rules: declining pages, unconverted organic traffic, high-impression low-CTR pages.
- Content groups and publication impact · compare page types, locales and topics; measure 14 days before and after a verified publish.
- Journeys and outcome cohorts · paths to the goal, assisted pages, signup → activation → purchase within a horizon.
- Page history · source diffs from GitHub, deploy markers, Search Console queries and observed AI citations in one page dialog.
- AI visibility · which AI crawlers read which pages, sessions referred by ChatGPT, Perplexity, Claude and Gemini, and citation checks.
Search · What changed 28 days vs previous
| page | clicks | Δ | from impressions | from CTR |
|---|---|---|---|---|
| /pricing | 1,204 | +18% | +31% | −13% |
| /blog/best-ai-video-models | 862 | −22% | −4% | −18% |
| /ai | 418 | new | — | — |
| /docs/getting-started | 396 | +6% | +9% | −3% |
Opportunities screening rules, not guesses
| page | rule | score |
|---|---|---|
| /blog/best-ai-video-models | clicks −22% on ≥ 20 prior | 0.81 |
| /compare/notion-ai | 1,400 imp · pos 6.2 · CTR 0.7% | 0.64 |
| /templates | 38 organic landings, 0 goals | 0.52 |
Product analytics that treats every visitor as a person.
Anonymous visitors get a profile from the first page view. Identify them later and history joins at query time, so events are never rewritten and merges are always reversible.
- Lifecycle graph · one picture from search impressions through acquisition to retention, with the paths between.
- Sequential funnels and retention · step k is the first occurrence after step k−1; windows, breakdowns, medians, weekly cohorts.
- Models · CHAID decision trees on leak-free datasets: who signs up, who comes back, as rules you can act on. Scores land in every profile.
- Companies · accounts from business-email domains and product memberships, with pricing, signup and activation signals per account.
- Session replay · opt-in rrweb recordings with masking, linked from the person and the event.
- Dashboards and SQL · saved dashboards with period deltas, CSV export everywhere, and a read-only SQL console on your own DuckDB file.
Models · Signup CHAID, temporal holdout · AUC 0.74
Retention weekly, signups by cohort
| cohort | w0 | w1 | w2 | w3 | w4 |
|---|---|---|---|---|---|
| Aug 4 | 100% | 42% | 31% | 27% | 25% |
| Aug 11 | 100% | 45% | 33% | 29% | — |
| Aug 18 | 100% | 51% | 38% | — | — |
From evidence to a launched campaign, with you steering.
Segments, email and webhooks are the customer data platform. The growth studio sits on top: it picks an opportunity from the analysis, drafts the creative for the channel, you adjust it, and the result is measured against a holdout.
- Segments · attribute rules, behavioural rules (did / did not, counts, sequences A → B) and groups, compiled to one SQL pass.
- Email campaigns and webhooks · frequency caps, one-click unsubscribe, signed webhooks or Slack on segment enter and exit, every send is an event.
- Tag manager and web personalization · remote click, scroll and time rules; on-site layers per audience, previewed on your real pages.
- Google Ads sets with economics · headlines, keywords, negatives and a budget projection you can tinker with, keyword by keyword.
- Banners, video, newsletters, blog posts · A/B variants, storyboards rendered to branded video, lifecycle emails, and articles published with one click.
- Brand guidelines and a decision log · voice, visual style and every dated decision feed the next generation. Adjust, redo or edit, never just approve or reject.
Campaigns · Paid acquisition from the "/pricing traffic, no signup" opportunity
Give routine work to a named AI employee. No card. Set up in 10 minutes.
Delegates tasks, reports back, works inside your tools.
Projection recomputes as you move the budget or exclude a keyword. Stop-loss and holdout are part of the plan, not an afterthought.
Built for the agent that runs your growth.
Every screen in cheese is also an MCP tool. Connect claude.ai, Claude Code, ChatGPT or your own agent with one URL and it can analyse, segment, draft campaigns and stage sends, inside the limits you set.
# claude.ai / ChatGPT / Claude Desktop: add a connector with this URL
https://cheese.teamday.cc/mcp
# Claude Code
claude mcp add --transport http cheese https://cheese.teamday.cc/mcp
# scripts and CI: a scoped key instead of OAuth
claude mcp add --transport http cheese https://cheese.teamday.cc/mcp \
--header "Authorization: Bearer sk_r_…"OAuth 2.1 with PKCE, so a connector needs nothing pasted. REST with OpenAPI 3.1 at /api/openapi.json for everything else.
66 tools, one workspace
Reports, funnels, retention, SQL, segments, campaigns, dashboards, models, site insights, page intelligence, companies, replays, tag rules, growth creatives.
Guardrails an agent cannot talk its way past
- Keys are scoped: read, write or admin. OAuth tokens are never admin.
- Real email sends and erasures need admin; campaign runs default to a dry run.
- Every tool call and mutating request lands in an audit log you can query.
- Ad-hoc SQL runs read-only in a sandboxed DuckDB instance with no filesystem access.
Live in two minutes.
No SDK to install, no data warehouse to fill, no tracking plan workshop. Page views, sessions and people appear on their own; add the events that matter to your business.
Paste one snippet
Before </head> on your site, or in Google Tag Manager. About 6.5 KB gzipped, batched, offline-safe, SPA aware, consent and Do-Not-Track aware, never throws.
Events become people
Every minute cheese rebuilds sessions, user vectors (any aggregate you declare, like sessions_7d or sum(props.amount)) and segment membership from your events.
Ask, segment, act
Open the workspace or connect an agent. Build an audience, attach an email or a webhook, generate the campaign, ship it, measure it against the holdout.
<script>
window.cheese=window.cheese||function(){(cheese.q=cheese.q||[]).push(arguments)};
cheese('init','wk_YOUR_WRITE_KEY',{host:'https://cheese.teamday.cc'});
</script><script async src="https://cheese.teamday.cc/v1/cheese.js"></script>
cheese.track('signup_completed', { plan: 'pro' })
cheese.identify('user_123', { email: 'ada@example.com', company: 'acme' })Fast on one server. No warehouse bill.
Each workspace is a single DuckDB file. Measured with 1,000,000 users and 100,000,000 synthetic events on one machine, so a marketing site of any realistic size is interactive.
All benchmark numbers (5 M and 100 M events)
| query | 5M events | 100M events |
|---|---|---|
| daily active users, 28d | 250 ms | 3.2 s |
| sessionize full history (window fn) | 3.9 s | 28.5 s |
| user vectors, 13 attributes, one pass | 1.8 s | 23.2 s |
segment sessions_7d ≥ 5 | 1 ms | 2 ms |
| funnel page → cta → signup, 28d | 322 ms | 3.7 s |
| weekly retention matrix, 8 weeks | 292 ms | 3.4 s |
| top paths by unique users, 7d | 186 ms | 2.2 s |
| revenue by plan via json_extract, 28d | 58 ms | 364 ms |
Synthetic data, DuckDB 1.5, 16 threads, 6.4 GB on disk. Live workspaces rebuild in milliseconds.
Your data, your file, our guardrails.
Built in the EU for teams that have to answer to a privacy officer. Nothing here is a setting you might forget to turn on.
EU-hosted, single tenant files
Every workspace is its own DuckDB file on a server in the EU. Isolation is a file boundary, not a row filter. Export the whole workspace as Parquet or the file itself, any time.
No raw IP addresses
Country is resolved at collection; only a salted hash of the address is kept. Consent, opt-out and Do-Not-Track are SDK calls, and replay masks input by default.
GDPR erasure that is real
One call removes a person from every table, tombstones the id so it cannot come back through a queue or a restore, and rewrites the raw log.
Clean numbers
Bots and AI crawlers are classified at ingest and excluded from every human report. Counts are labelled observations; zero is never reported as proof of zero.
Scoped keys and an audit trail
Read, write and admin keys, hashed at rest, with per-key rate limits. Every mutation, tool call, erasure and real send is logged with who, what and when.
Hardened by default
Strict per-request CSP, sandboxed SQL, signed webhooks, one-click unsubscribe and suppression lists, daily backups with a restore path that is tested.
Pricing that stays flat when your traffic doesn't.
Priced per workspace, not per seat and not per tracked person. Every plan includes analytics, the customer data platform and agent access.
Free
For a marketing site or a side project finding its footing.
- 50,000 events / month, 1 workspace
- Analytics, funnels, retention, dashboards, SQL
- Segments, email campaigns, webhooks
- MCP and REST with scoped keys
- Community support
Pro
For a product team that wants the whole picture without a data stack.
- 1,000,000 events / month, 3 workspaces
- Everything in Free
- Site intelligence with Search Console
- Session replay and companies
- Models, page history, AI visibility
- Email support
Business
For a growth team that ships campaigns every week.
- 10,000,000 events / month, 10 workspaces
- Everything in Pro
- Growth studio: ads, banners, email, blog posts
- Branded video generation
- Tag manager and web personalization
- Priority support, onboarding call
Self-hosted
For companies whose data may not leave their own infrastructure.
- The whole product on your server
- One box, no other infrastructure
- Unlimited events and workspaces
- Your own AI provider keys
- Support agreement
Events are what your site sends: page views, clicks, product actions. Bot traffic is filtered before it counts. Need more events on a plan? Ask, it is a number in a file.
Questions we get asked.
Is this a replacement for Google Analytics?
For a product or marketing site, yes, and it goes further: people instead of sessions, segments you can email, and Search Console data joined to your own goals. It does not replace Google Ads or Search Console themselves; it reads from them.
How is it different from Mixpanel or PostHog?
They stop at analytics or at a feature list. cheese connects analytics to the customer data platform and to campaign generation, and exposes all of it to AI agents with scoped permissions. It runs on one server with one file per workspace, which is why the price is flat.
Do I need a data warehouse or an engineer?
No. One snippet, and page views, sessions, sources and people appear. Product events are one line of code each. Attributes, segments and dashboards are declared in the UI or by an agent, never migrated.
What can an AI agent actually do with it?
Read every report and run SQL, build and preview segments, stage campaigns and dashboards, draft ads and emails, and inspect what changed on a page. Sending real email, erasing a person or managing keys needs an admin key that you never give an agent.
Where is my data and can I take it with me?
On a server in the EU, in a DuckDB file that belongs to your workspace. Export it as Parquet or download the file. Erase a person and they are gone from every table, the raw log and every future restore.
Does the growth studio publish things on its own?
Never. It drafts from evidence and your brand guidelines, you adjust or edit, and only then does a campaign go to email, a tag rule go live or a post get published. Every decision you make is remembered for the next draft.
One workspace. One file you own.
Create a workspace, paste the snippet, connect your agent. See your first visitor in two minutes, free while in beta.
Start free