Built for the DataHub Agent Hackathon · Agents That Do Real Work

A table just broke.
Ripple already knows what's on fire.

Ripple is an AI agent for DataHub. It traces the downstream blast radius of a broken asset, ranks the impact by severity, and writes the incident back into the catalog — so the next engineer inherits the answer.

Try it — simulate an incident

Click any table to break it. Ripple's logic runs right here in your browser — no install, no DataHub — tracing the blast radius, ranking severity, and writing the incident back.

Tip: break orders_raw for the full blast radius, or a leaf like Executive Revenue to see it contained.

Not just a query tool — an agent that acts

Ripple reads lineage, reasons about impact, and writes the incident back to DataHub. Everything below works end-to-end.

Impact

Downstream blast radius

Trace every affected table and dashboard across all lineage hops, ranked by criticality with an auto-assigned SEV level.

Root cause

Upstream analysis

Flip the direction: trace upstream to rank the likely sources of bad data, closest raw tables first.

Precision

Column-level lineage

Not just "this dashboard" — the exact column that traces back to the broken field.

Autonomy

Auto-trigger watch

Poll for broken assets and triage them automatically — no human in the loop to kick it off.

Write-back

Native incidents

Applies an incident tag, saves a runbook, and raises a first-class DataHub Incident entity on the asset.

Interface

Web + terminal

A rich CLI and a read-only web dashboard with an interactive lineage graph and light/dark themes.

How it works

One command: read the lineage, reason about impact, write the incident back.

1

Read

Traverse downstream lineage across every hop and platform via the DataHub MCP Server — Snowflake, dbt, Looker, PowerBI, Tableau.

2

Reason

Rank the blast radius by criticality and assign a severity. Facts are gathered by code; only the narrative is written by an LLM — so the analysis stays trustworthy.

3

Write

✓ incident tag

✓ runbook document

✓ native Incident entity

Persisted back so the next engineer inherits full context.

Run it in one command

Works against any DataHub instance — a local quickstart is enough.

# install
uv pip install -e .
# triage a broken asset
python -m ripple triage "urn:li:dataset:(...,orders_raw,PROD)"