dltHub
Book a demo

Agents that debug your pipelines, within the guardrails you set

Delegate the maintenance work, keep your engineers in charge, and bring data agents into production one permission at a time.

Book a demoRead the docs

Build your first dltHub background agent

Copy the full setup prompt and paste it into your coding agent to add an agent to your dltHub workspace.

Add job-inspector to my workspace and run it when the load fails

Your engineers build ten times more pipelines. Now agents can diagnose them too.

Finding the cause of a failed job used to be the first stretch of every incident.
Now it's done shortly after the pipeline failed.

Root cause analysis in your inbox

Accelerate the fix by a ready-made diagnosis.

Self-improving pipeline health

Teach the agent to get better with each failure.

The agent is declared like any other job

Trigger it on a success, a failure or a schedule.

Declare the agent job
__deployment__.py
@run.agent(    agent="dlthub-platform:job-inspector",    trigger=[job_inspector.fail],    execute={"concurrency": 10},)

Less time spent on incidents, more time available for insights

A data engineer opens the diagnosis of the root cause with a proposed fix, the agent's confidence level and the evidence attached. The repair is only one prompt away.

Job fails

Job Runs
JobStartedStatus

load_stripe_payouts

stripe_payouts

2m ago

Agent investigates

Agentsjob_inspector
  1. Done:logs1.1s
  2. In progress:diagnosiswriting…

Report lands

Slack1

Coding agent fixes

Claude CodeCodexCursor

You know why the pipeline failed before you open the logs

dltHub’s verified job-inspector agent automatically diagnoses your pipelines after a failure. Every failure comes back with the likely cause, proposed fix and the evidence to check it.

The agent is a debugging expert

Docs

The background agent analyzes the pipeline trace, the logs and the job definition together. It knows dltHub inside and out and is trained on hundreds of production failures.

#data-alerts

Slack

Job InspectorAPP2:06 AM

load_stripe_payouts #4127 failed · diagnosis ready

Likely cause

The source now sends total_amount as a decimal. The job expects an integer, so the load stopped at the schema check.

Confidence

High

Diagnosis based on evidence

Docs

The agent reports its confidence level and the evidence it found. A data engineer can hand over its coding agent a direction rather than starting a search.

Diagnosis

  • The load step failed for resource payouts: total_amount expected BIGINT, got DECIMAL.
  • The source now sends total_amount as a decimal, and the schema expects an integer.

Recommendation

  • Coerce total_amount in the payouts resource, or update its schema hint and redeploy.

Explore the rest of dltHub