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

# Pipeline

> Core concepts surrounding Arivu's LangGraph compiled nodes.

Arivu's intelligent query resolving utilizes a sophisticated LangGraph pipeline that abstracts schema ingestion, LLM execution, error verification, and memory logging. The system is designed for granular observability and deterministic human feedback.

<Info>The pipeline is the core engine that transforms natural language into secure, verified SQL operations.</Info>

## Pipeline Lifecycle

<Steps>
  <Step title="Query Intake" icon="input">
    Capture raw user intent with metadata context (database engines, session memory, user context).
  </Step>

  <Step title="SQL Generation" icon="code">
    Convert the parsed intent using LLMs with knowledge of exact relational schemas.
  </Step>

  <Step title="Verification" icon="shield-check">
    Evaluate generated SQL for:

    * **Syntax validity** - Is the SQL correct?
    * **Safety checks** - Blocks destructive operations (DROP, DELETE, etc.)
    * **Schema validation** - Prevents hallucinated columns/tables
  </Step>

  <Step title="Execution & Memory" icon="database">
    Execute approved queries and sync results to the internal memory system for session context.
  </Step>
</Steps>

## Graph Inspection

Arivu compiles the pipeline into a LangGraph for transparent execution tracking. Inspect it locally:

<CodeGroup>
  ```python inspect_graph.py theme={null}
  from arivu.pipeline import get_compiled_graph

  graph = get_compiled_graph()

  # Print ASCII visualization
  print(graph.get_graph().draw_ascii())

  # Export as Mermaid diagram
  print(graph.get_graph().draw_mermaid())
  ```
</CodeGroup>

## Key Concepts

<AccordionGroup>
  <Accordion title="State Management" defaultOpen icon="layers">
    The pipeline maintains a shared state dictionary that flows through each node:

    * User query and session context
    * Generated SQL candidates
    * Verification results
    * Final response and metadata
  </Accordion>

  <Accordion title="Error Handling" icon="alert-circle">
    When verification fails, the pipeline:

    1. Captures the error
    2. Provides feedback to the LLM
    3. Automatically attempts correction
    4. Re-verifies the corrected SQL
  </Accordion>

  <Accordion title="Observability" icon="activity">
    Every pipeline execution is tracked in your memory backend:

    * Timing metrics for each node
    * Token usage for LLM calls
    * Full state history for debugging
    * User session context preservation
  </Accordion>
</AccordionGroup>

<Tip>Use the Arivu Dashboard to visualize pipeline execution traces and performance metrics in real-time.</Tip>