Too many tabs
Logs in one window. SQL in another. Pipeline settings somewhere else. The evidence is scattered across tools that do not talk to each other.
An AI data engineer for failed pipelines
Your pipeline failed. Here’s the fix.
PipelineMender AI investigates failed runs on Microsoft Fabric, Azure Data Factory and Databricks, connects the evidence, and hands you a specific fix to review.
- Read-only access
- You review every change
- No model training
daily_orders_ingest
Pipeline failed
Investigating
Fix ready for review
- 01Failure
- 02Investigation
- 03Evidence
- 04Root cause
- 05Fix
01:58:12Copy_Orders failed · SqlOperationFailed
01:58:12Invalid column name 'customer_id' (Msg 207)
01:58:40Reading output, definition, source schema
01:58:52Schema: client_id present · customer_id absent
- Logs
- Schema
- Source query
- 3 sources agree
Likely root cause·confidence strong
customer_id was renamed to client_id in the source. The Copy activity still maps the old name.
Recommended source querySQL DIFF
01SELECT order_id,02- customer_id,03+ client_id AS customer_id,04 total_amount, updated_at05FROM dbo.Orders;
18,420 records delayed
No changes applied · review in your tenant
Built for
- FMicrosoft Fabric
- AAzure Data Factory
- DDatabricks
- Schema drift
- Expired credentials
- Missing input files
- Source timeouts
- Dependent task failures
- Type mismatches
- Configuration errors
- Permission changes
The alert is only the beginning
2 AM pipeline failures cost hours.
A failed activity tells you where the run stopped. You still need to find the cause, trace the impact, and work out what to change.
Unclear impact
A failed run can leave downstream tables and reports stale. Knowing how many records are delayed decides how urgent the fix is.
A fix still to find
Retrying the same configuration does not repair a changed schema. Someone has to read the error, compare the schema, and write the change.
From failure to a reviewed fix
One investigation. Every step connected.
Every 5 minutes PipelineMender syncs your connections. A failed run becomes an incident, and the investigation starts on its own.
-
01
Failure
A failed run is detected and opened as an incident with its run context.
-
02
Investigation
Logs, activity output, pipeline definition, schema and permitted data are read.
-
03
Evidence
Each finding is kept as a numbered piece of evidence you can inspect.
-
04
Root cause
The likely cause, its confidence, and the evidence that supports it.
-
05
Fix
A specific SQL, code, configuration or credential change, ready for your review.
Example investigation
One renamed column.
A blocked orders pipeline.
The source schema changed. The Copy activity did not. PipelineMender connects the error to the schema change, shows the evidence behind it, and proposes an alias that preserves your destination mapping.
- ObservedSQL error 207 for
customer_id. - Evidence
client_idexists; the old column is absent. - Impact18,420 records delayed; 0 rows written.
Every new workspace starts with this sample incident, so you can read a full report before connecting anything.
SeverityHigh
Records delayed18,420
ConfidenceStrong
Root cause
The source table dbo.Orders no longer has a customer_id column; the same data is now in client_id. Copy_Orders still selects and maps customer_id, so the source query fails before any rows are read.
E1 LogsE2 SchemaE3 Source query
Evidence
- E1Activity outputMsg 207, Level 16: Invalid column name 'customer_id'.
- E2Source schema
dbo.Ordershasclient_id;customer_idis absent. - E3Source query
sqlReaderQueryand the column mapping referencecustomer_id.
Recommended fixReview before applying
01SELECT order_id,02- customer_id,03+ client_id AS customer_id,04 total_amount, updated_at05FROM dbo.Orders06WHERE updated_at >= '@{pipeline().parameters.window_start}'07 AND updated_at < '@{pipeline().parameters.window_end}';
Aliasing keeps the destination mapping unchanged.Was this report right? YesPartlyNo
Your stack, connected
Start where your pipelines run.
Connect with the credentials your platform already issues. Every connection is read-only, and Test tells you exactly which role is missing before the first sync.
FRead-only
Microsoft Fabric
Workspace pipelines, activity runs and pipeline definitions.
- Sign-in
- Entra app registration
- Role
- Workspace Viewer
ARead-only
Azure Data Factory
Factory pipeline runs, activity runs and pipeline configuration.
- Sign-in
- Entra app registration
- Role
- Reader on the factory
DRead-only
Databricks
Jobs, tasks and notebook runs, including Free Edition workspaces.
- Sign-in
- Access token or service principal
- Role
- Can view on jobs
Security & trust
Investigate with access you control.
Read-only by design
PipelineMender reads logs, metadata and the data you authorize. It never executes the recommended fix; you apply it in your own environment.
Microsoft Entra ID
Sign in with your work account, or with e-mail and a one-time code. Organization access stays tied to your identity provider.
No model training
Your pipeline data and your feedback on reports are never used to train models.
Retention you control
Reports and evidence are purged on your schedule, from 7 to 365 days. Record sampling is off by default and scoped per connection.
Monthly billing
Start small. Add your team.
Choose coverage for one workspace or a shared data platform.
Starter
$49/ month
For one engineer and one workspace.
- 1 connection · 1 user
- 50 investigations per month
- 7-day report retention
TeamFor shared platforms
$249/ month
For engineers sharing on-call work.
- 5 connections · 10 users
- 500 investigations per month
- 30-day reports · shared ownership
Enterprise
Custom/ month
For organization-wide coverage.
- Negotiated connections and seats
- Custom retention and deployment review
- Procurement and support options
Prices in USD. Trial length, overages and final terms will be confirmed at launch. Every plan includes Microsoft Entra ID sign-in and read-only access.
Before you connect
Questions engineers ask.
Does PipelineMender change my pipelines?
No. Connections are read-only. Review the recommended change, apply it in your own environment, and validate the next run.
Which platforms are supported?
Microsoft Fabric, Azure Data Factory and Databricks (including the free edition) are supported at launch.
What can it read?
Only the workspaces, factories, metadata, logs and source data you authorize. Start with metadata and logs; relevant record access is separately scoped and off by default.
What if the root cause is uncertain?
The report separates observed facts from hypotheses, shows missing evidence, and tells you what to check next. It does not invent a fix when evidence is insufficient.
Is our data used to train models?
No. Your data is not used for model training. Retention and authorized data access are visible in organization settings.
How does billing work?
Plans are billed monthly. Investigation allowances, overages and final prices will be confirmed before launch.
From failed run to reviewed fix