Skip to main content
oleander includes two assistants - Lea and Ollie - that can investigate pipeline failures, explain cost spikes, trace lineage, and query your lake directly, without you writing SQL or navigating dashboards.

​
What they can do

Lea and Ollie have access to your full context graph through a built-in set of tools. They can:
  • Investigate a failed pipeline run and produce a root cause summary with evidence
  • Trace lineage for a dataset or job upstream and downstream
  • Query the lake with SQL and return results inline
  • Explain what a Spark job did - inputs, outputs, schema changes
  • Surface the slowest jobs or most expensive pipelines
  • List recent failed runs and summarize error patterns
  • Compare two pipeline runs and highlight differences
  • Show dataset change history - what schema changed and when

​
How to use it

Open the chat interface in the oleander platform and ask in plain English. You do not need to know run IDs, namespace names, or table paths - the assistants can look those up. Example prompts:
  • “Why did last night’s ingestion job fail?”
  • “What tables does the orders_daily job write to, and who reads them?”
  • “Which job is costing the most this week?”
  • “Show me the schema of oleander.default.customers and when it last changed.”
  • “Did any jobs fail in the last 24 hours? Summarize the errors.”
  • “Compare the last two runs of transform_events - what was different?”

​
Investigations

When a pipeline fails, oleander can trigger a full automated investigation. It gathers logs, OTel spans, lineage events, Iceberg commit metrics, and Spark plan diffs, then produces:
  • A root cause finding (1-4 sentences)
  • Evidence cited from telemetry and lineage
  • Recommended next steps
Investigations can be triggered from the alert detail view or by asking Lea directly. You can retrieve past investigation results through the MCP using oleander_list_investigations and oleander_get_investigation.

​
Personas

Lea focuses on compliance, governance, and data quality - useful for questions about dataset ownership, schema contracts, and downstream impact of changes. Ollie focuses on pipeline operations, cost, and performance - useful for incident investigation, cost attribution, and optimization. Both have access to the same underlying tools. Switch personas from the chat interface.

​
Programmatic access

For agent-driven workflows, the same investigation context is available through the MCP tools and API. Use oleander_list_investigations to retrieve past investigations or trigger new ones from your agent.