> ## Documentation Index
> Fetch the complete documentation index at: https://docs.delphina.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Research Agent

> Deep, multi-step investigations with written reports

The research agent runs a multi-step investigation. The agent creates a plan, executes multiple queries, tests hypotheses, and produces a written report with findings and next steps.

Use it when you need to understand *why* something is happening, not just *what* the numbers are.

## When to use it

* Broad or strategic questions requiring multiple queries
* You want the agent to propose metrics, validate them, and write up findings
* Questions like "Why did Q4 revenue spike in week 3?" or "What factors are driving churn?"

For quick factual questions ("What was revenue last month?"), use the default [Analytics Agent](/analytics-agent/overview).

## How to start

Start a **new chat** with `/research` followed by your question:

```text theme={null}
/research analyze churn drivers for enterprise customers over the last 6 months
```

<Note>
  `/research` can only be used on the **first turn** of a chat.
</Note>

## What the agent does

**1. Planning.** Searches your knowledge base, proposes metrics and SQL, and creates a structured plan with hypotheses.

**2. Research.** Queries your warehouse to stage data, then uses Python and DuckDB in a secure sandbox to analyze, test hypotheses, and build charts. Each step builds on the last — the agent can query, inspect results, refine its approach, and query again. All execution happens in an isolated container with no internet access.

## Output

* A **plan summary** of the question and approach
* **Hypotheses** with statuses: open, supported, or rejected
* **Findings** tied to specific SQL results and charts
* **Next steps** with concrete follow-up questions

## Getting better results

Include time ranges ("last 6 months"), segments ("enterprise customers"), and specific metrics in your question. Document your knowledge base first — the research agent produces better results with documented table schemas and metric definitions.