FOR DATA ENGINEERS

Debug the payload
before it breaks the DAG.

Event streams, webhook payloads, dead-letter records, config blobs — inspect the one malformed message, count events by type, or catch schema drift by asking. QDev writes the jq and runs it locally.

✓Find the record with the null field without a scratch script
✓Counts & rollups are exact — jq computes them, not a model
✓Production data & PII stay on your machine — never uploaded

Add to Chrome Add to Firefox

⌘K

try

● EVENTS.NDJSON — 128,400 RECORDS · LOCAL

{
  "events": [
    {
      "ts": "2026-07-04T09:14:02Z",
      "type": "page_view",
      "user_id": "u_1204",
      "status": "ok"
    },
    {
      "ts": "2026-07-04T09:14:03Z",
      "type": "purchase",
      "user_id": null,
      "status": "ok"
    },
    {
      "ts": "2026-07-04T09:14:05Z",
      "type": "checkout",
      "user_id": "u_0098",
      "status": "error"
    }
  ]
}

↑ Ask a rollup or a validation — jq scans all 128k records locally. Only the schema is ever sent.

exact rollups · reproducible jq · nothing uploaded

SOUND FAMILIAR?

3am, the pipeline's red

Schema drift, silently

An upstream service started sending user_id: null and half a partition got dropped before anyone noticed.

The dead-letter queue

A few thousand records failed to parse. You need the one field that's wrong across all of them — fast.

NDJSON too big to grep

A 300 MB export won't open, and jq on the CLI means quoting a filter three times to get the escaping right.

ASK, DON'T SSH INTO A NOTEBOOK

From a bad record to a root cause

🧬

Catch schema drift

Find records missing a field, with a wrong type, or a value outside the enum you expected.

events where status isn't ok or error

📈

Roll up & count

Events by type, error rate per source, volume per hour — exact counts, computed by jq.

count events by type per hour

🧯

Triage the DLT

Isolate the malformed records and see exactly which field failed, across the whole batch.

records where amount is not a number

🪆

Flatten nested records

Explode a nested payload into flat rows ready to reload or diff against the source.

flatten payload.items into one row each

FITS YOUR STACK

Wherever your JSON lands

STREAMS

Kafka / events

Event and message payloads — grouped, counted, and validated locally.

NDJSON

Line-delimited

Log exports and batch files too big to open elsewhere.

WEBHOOKS

API payloads

Inspect the exact body that failed a downstream contract.

CONFIG

Configs & manifests

dbt, Airflow, package manifests — read the shape at a glance.

# QDev: "records where user_id is null" → paste the jq into your DLT replay [.events[] | select(.user_id == null)] | length # 412 dropped

● PROD DATA STAYS IN PROD

Payloads never leave the tab.

Customer records, PII, tokens caught in an event — the values stay local. Only the schema (field names & types) and your question are ever sent, and only on the hosted tier. Handling regulated data? Run local AI, free, and nothing leaves at all.

✓jq runs in a sandboxed Web Worker — no network access

✓Hosted proxy is stateless & zero-retention — logs nothing

✓No data pipeline to a vendor, no sampling, no copy of your stream

One record is malformed. Find it.

Install free, keep your stream local, and let ⌘K write the jq that isolates the bad payload.