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

# Chat

> Ask questions, get suggestions, and troubleshoot pipelines with AI chat.

The **Copilot chat** panel is your primary place to ask how something works, sketch a design, or debug an error message. Conversations stay beside the editor so you can apply suggestions without switching tools.

## Sidebar chat interface

Open **AI Copilot** from the pipeline toolbar or global assistant entry point. The sidebar shows:

* Your messages and the assistant replies
* Optional **context chips** for the active pipeline, node, or error
* Shortcuts to insert generated SQL or node snippets when the model offers them

Resize the panel or collapse it when you need full canvas width. Chat history is scoped per workspace according to your retention policy.

<Tip>
  Start messages with the outcome you want (“I need incremental loads keyed on `updated_at`”) and name the systems involved (S3, BigQuery, Salesforce). Specific prompts reduce back-and-forth.
</Tip>

## Pipeline context-aware suggestions

When you work inside a pipeline, Copilot can read **structural context** you authorize: node types, edge relationships, and configuration fields needed to give relevant answers. It does not automatically exfiltrate full row-level data from previews unless you explicitly paste samples.

<Tabs>
  <Tab title="Node-focused questions">
    Select a node and ask about its parameters, performance, or error text. The assistant ties answers to that node’s type (source, transform, destination).
  </Tab>

  <Tab title="End-to-end flow">
    Ask how data moves from source to destination; Copilot summarizes the graph and suggests missing quality checks or orchestration hooks.
  </Tab>

  <Tab title="Run failures">
    Paste or link run IDs when your role allows; Copilot explains likely causes and points to logs or [diagnostics](/observability/diagnostics).
  </Tab>
</Tabs>

You can clear context between topics so unrelated pipelines do not leak into the next answer.

## Feedback system

Use **thumbs up / thumbs down** (or the feedback form, if enabled) on replies your organization uses for quality review. Feedback helps admins decide when to adjust prompts, update the [knowledge base](/ai-copilot/knowledge-base), or restrict certain behaviors.

<AccordionGroup>
  <Accordion title="What to include in feedback">
    Note whether the answer was factually wrong, unsafe, or merely verbose. Mention the pipeline type and node if the issue was context-specific.
  </Accordion>

  <Accordion title="Escalation">
    For repeated hallucinations or policy concerns, open a ticket with your internal AI governance contact; do not rely on chat for compliance sign-off.
  </Accordion>
</AccordionGroup>

## Limitations

Copilot may propose deprecated parameters or connector options that changed in your workspace version. Always verify against the in-product docs and run a preview. If chat is disabled, you still have full manual authoring capabilities.

## Related topics

<CardGroup cols={2}>
  <Card title="Pipeline generation" icon="diagram-project" href="/ai-copilot/pipeline-generation">
    Turn descriptions into node graphs.
  </Card>

  <Card title="SQL assistance" icon="database" href="/ai-copilot/sql-assistance">
    Generate and tune queries inside nodes.
  </Card>
</CardGroup>