You already have a data science expert.
Your coding agent.

Instrument once. Ask better questions.

Keep your existing coding agent. Give it what your users do and say, and let it act on it.

What your agent hands back.

Real reports built by Slurp from sample data, through the same pipeline you would use: line and bar charts, heatmaps, donuts, scatter plots, funnels and areas, in any Perspective theme, with retention, funnels, visits and next-step paths from DuckDB's behavioral functions. Open one: each is an ordinary self-contained HTML file.

Why export completion droppedLine, grouped bars, tableDo new users come back?Cohort heatmap, retention curvesHow fast do new users get value?window_funnel, barsWhat happens after a failure?sequence_match, next stepHow engaged are people?sessionize, lines, barsWhat are people exporting?Donut, bars, scatterWhere does the flow leak?Funnel, bars, areaHow did each release do?Stacked bars, area, tableThe same report, Pro DarkAny Perspective theme

Sample data from a fictional app, generated by examples/demo and loaded through a real Worker. Nothing here is customer data.

Your agent writes the report in Markdown.

A report is a Markdown file with named SQL and a few blocks. The query stays with the finding, so every number can be reproduced. This is the whole source of the chart beside it.

by-release.mdCopy ↗

---
purpose: Which release broke exports
---

# Which release broke exports?

```sql by_release
SELECT
  context->>'$.app.version' AS release,
  round(100.0 * count(*) FILTER (
    WHERE event = 'export_completed'
  ) / count(*) FILTER (
    WHERE event = 'export_started'
  ), 1) AS completed
FROM events
GROUP BY 1
ORDER BY 1
```

```chart
query: by_release
title: Exports completed, % of started
spec:
  height: 170
  mark: bar
  encoding:
    x:
      field: release
      axis: {labelAngle: 0}
      title: null
    y:
      field: completed
      type: quantitative
      title: null
    color:
      field: release
      legend: null
```

• Markdown is the source of truth. Diffable, reviewable, sanitized.
• HTML is the main output. One self-contained file with its charts and fonts.
• PNG and PDF come from the same source. Ready for Slack, email and slides.
• Your agent already knows how. slurp agent setup installs the skill.

slurp report check report.md
slurp report build report.md --out ./report --png --pdf
slurp report open ./report

A few lines. Real events.

A publishable key sends events. Your read key brings them home.

import { init, track, flush } from '@slurp/sdk';

init({
  endpoint: 'YOUR_INGEST_URL',
  sourceId: 'YOUR_SOURCE_ID',
  key: 'YOUR_PUBLISHABLE_KEY',
});

track('signed_up', { plan: 'pro' });
await flush(); // Node: flush before exiting

The SDK is private until published. The quickstart includes installation from source.

Bring more of your stack.

Start with SDK capture, HTTP, and your coding agent. Explore what could come next.

JavaScript / TypeScript

Browser and Node event capture. Install the private SDK from source.

Setup guide ↗

Available locally

HTTP API

Send canonical events as NDJSON to your configured backend.

Setup guide ↗

Works today

Stripe

Payments and subscriptions.

Exploring

GitHub

Pushes, pull requests, and releases.

Exploring

Linear

Issues and cycles.

Exploring

Sentry

Errors and releases.

Exploring

Your data. Your machine. Your agent.