Kepler turns raw product events into funnels, cohorts and session replays in seconds — warehouse-native, genuinely self-serve, and priced on what you use instead of who logs in.
Measuring what matters at
Most analytics stops at the chart. Kepler keeps going — from the number, to the cohort behind it, to the session that explains it, without leaving the page.
Build a funnel from any five events, then click the biggest fall to see exactly who stalled and what they did instead.
Weekly cohorts side by side, so you can see whether last month's onboarding change actually bent the curve.
-- no export, no sync, no copy of your data select plan, count(distinct team_id) as teams from kepler.events where name = 'activated' and ts > now() - interval '30 day' group by 1 order by 2 desc;
Kepler reads the same tables your BI stack does. One definition of "activated" — the dashboard and the board deck finally match.
Anomaly detection on any metric, delivered to Slack with the chart already attached.
Click a drop-off in a funnel and Kepler queues the replays of the people who fell out of it — with console errors, rage clicks and network failures already marked on the timeline.
Watch a sample replayNo data team required, no six-week implementation, no professional services invoice. Paste one snippet and start asking questions.
One snippet, or point Kepler at the warehouse tables you already have. Autocapture fills the gaps while you define the events that matter.
$ npm i @kepler/analytics
Name your events in plain language and mark the three that matter. Everyone in the workspace inherits the same definitions — no duplicate "signup_v2".
› kepler.track('activated')
Funnels, cohorts, paths and replays from the same event stream — and a SQL escape hatch for the day the UI isn't enough.
? why did activation drop
Bad data is not a reporting problem, it's a schema problem. Kepler validates every event against your spec on the way in and quarantines what doesn't match — so nobody discovers a broken funnel three weeks later.
Fix the spec, hit replay — history is corrected in place.
A dashboard nobody can change is a dashboard nobody trusts. In Kepler anyone can fork a chart, annotate a spike, and hand the question back with context attached.
"We replaced two tools and a standing Tuesday meeting. The thing I did not expect: our PMs stopped filing data requests — they just answer their own questions now."
"Warehouse-native was the whole decision. Finance and product finally quote the same number, which had never once been true before."
"Onboarding took an afternoon. Our old tool took a quarter and a professional services contract we're still paying for."
"The replay-from-funnel jump is the feature I demo to everyone. Ten seconds from 'conversion dropped' to watching it happen."
"Usage pricing meant we could give every engineer an account on day one. That changed how the whole company argues about product."
"Schema validation caught a broken checkout event the same hour it shipped. Previously that would have been a month of bad reporting."
Invite the whole company — support, design, the founder who keeps asking. You only pay for the data you send.
For the first product and the first thousand users.
$0
Free forever, up to 1M events / month
For teams shipping weekly and measuring every release.
$89 / month, billed monthly
Includes 20M events, then $4 / million
For regulated teams and very large event volumes.
Custom
Annual contract, volume pricing
All plans include unlimited seats, unlimited dashboards and the full API. Startups under two years old get 12 months of Scale free.
No. Kepler can run entirely on your warehouse — Snowflake, BigQuery, Redshift or ClickHouse — reading the tables you already maintain. If you would rather we host the storage, we can do that too, in the EU or the US. The product is identical either way.
On events ingested per month. Every plan includes unlimited seats, dashboards and API access. We think charging per viewer is a tax on the exact behaviour we want — more people looking at the data — so we don't do it.
Point Kepler at your existing event stream and run both in parallel for a fortnight; we backfill history from your warehouse or from an export. Most teams cut over in under three weeks, and we'll do the schema mapping with you at no cost.
Yes. Field-level redaction happens before anything is persisted, so flagged PII never reaches storage. We are SOC 2 Type II certified, support EU-only residency, and offer a DPA with standard contractual clauses on every paid plan.
Median query time is about 400ms across a billion-row dataset, and p95 stays under two seconds. Funnels and retention are computed with pre-aggregated sketches, so widening a date range doesn't linearly widen the wait.
On the Enterprise plan, yes — Kepler ships as a Helm chart that runs in your own VPC, with the control plane still managed by us for upgrades. Fully air-gapped installations are available for regulated customers.
Free up to a million events a month, no card, no sales call, no implementation project. If it isn't answering real questions by Friday, delete the workspace.