Skip to content

Quickstart

Verify the detection logic offline in thirty seconds, then drive a live collector.

Molt has two separate on-ramps, and they are worth telling apart before you start.

The first needs no credentials and proves the detection logic is sound. The second drives a real collector against Bright Data Scraper Studio and costs credits. Most people should do the first one first.

Verify the logic, no account needed

Shell
pnpm install
pnpm test

This runs the full suite offline against committed fixtures of real Bright Data output — including the drift-detection core, the heal-prompt generator, and the whole incident state machine. Nothing here touches the network or reads an API key.

Note

CI runs this exact command twice: once normally, and once with every credential blanked (BRIGHTDATA_API_KEY=''). The second run exists to prove the claim above is actually true, not just stated.

Drive a live collector

Shell
npx -p @brightdata/cli bdata login
Shell
cp .env.example .env

Paste your collector IDs into .env, then:

Shell
pnpm molt init
Shell
pnpm molt check primary

molt init registers the collectors your .env points at. molt check runs one of them through the real Bright Data CLI, snapshots the result, and compares it to whatever baseline exists.

Warning

check, watch, approve, and add all shell out to the real bdata CLI and can spend credits or take minutes to return — heal and create in particular are AI-Flow jobs. See Honest limits before running them against an account you care about.

The shape of a session

Shell
pnpm molt check chaos
Shell
pnpm molt watch
Shell
pnpm molt review
Shell
pnpm molt approve

Check for drift, advance whatever incident that opens as far as it can go automatically, review the proposed fix as a human, then approve it — which also re-runs the collector to verify the fix actually worked. Heal & review walks through what happens at each of those four steps and why the third one cannot be automated away.

Where to go next

  • Concepts — snapshot, baseline, drift, incident: the five nouns everything else is built from.
  • molt check — what one run actually does, and how a verdict is produced.
  • CLI reference — every command, in one table.