AI Suggestions

VectorFlow includes optional AI-powered assistance for writing VRL code and generating pipeline configurations. When enabled, team members with Editor or Admin roles can use AI features in both the VRL editor and pipeline builder.

Setup

Team admins can configure AI in Settings → AI. The configuration requires:

After saving, use Test Connection to verify the configuration works.

VRL Assistant

In the VRL editor (opened from any remap, filter, or route transform), click the AI button in the tools panel to reveal the AI input.

1. Type a natural language description of what you want the VRL code to do
2. Click Generate — the AI streams VRL code in real time
3. When complete, choose:
   • Insert — append the generated code after your existing code
   • Replace — replace all existing code with the generated result
   • Regenerate — try again with the same prompt

The AI is aware of your upstream source types and available fields, so you can reference them naturally (e.g., "parse the syslog message and extract the hostname").

Pipeline Builder

In the pipeline editor toolbar, click the sparkle icon to open the AI Pipeline Builder dialog.

Generate mode

Describe a pipeline in plain language:

"Collect Kubernetes logs from a file source, drop debug-level events, parse JSON, and send to Elasticsearch and S3"

The AI generates a complete Vector YAML configuration. Click Apply to Canvas to add the generated components to your pipeline. If your canvas already has components, the new ones are positioned below the existing layout to avoid overlap.

Review mode

Ask the AI to analyze your current pipeline configuration:

"Are there any performance issues with my pipeline?"

The AI returns structured, actionable suggestion cards that you can selectively apply to your canvas.

Suggestion cards

Each suggestion appears as an interactive card showing:

• Title and description explaining why the change helps
• Priority badge (High, Medium, Low)
• Type badge — Config Change, Add Component, Remove Component, or Rewire
• Checkbox for batch selection
• Config preview for configuration changes showing the exact fields that will be modified

Applying suggestions

• Apply All — applies every actionable suggestion from that AI response
• Apply Selected — applies only the suggestions you have checked

Applied suggestions are marked with a green "Applied" badge and cannot be re-applied. The entire batch is a single undo operation — press Ctrl+Z (or Cmd+Z) to revert all changes at once.

Conflict detection

When you select multiple suggestions that conflict (e.g., two suggestions modifying the same config field, or one removing a component that another references), an amber warning appears on the affected cards explaining the conflict. You can still apply conflicting suggestions, but review the warnings first.

Suggestion statuses

Conversations

Review conversations are persistent — they are saved per pipeline and visible to all team members with access. You can:

• Ask follow-up questions using the input at the bottom of the dialog
• Start a new conversation by clicking "New Conversation" below the input

Rate Limits

AI requests are rate-limited to 60 requests per hour per team to prevent excessive API usage. The limit resets on a rolling window.

Security

• API keys are encrypted at rest using AES-256-GCM
• Keys are never exposed to the client — the settings page shows only whether a key is saved
• AI configuration changes are recorded in the audit log with the API key redacted
• Only team members with Editor or Admin roles can use AI features