Lagniappe can use Google Vertex AI for optional generation, extraction, and summarization features. AI runs only when a user invokes an AI action.

The goal is not to sprinkle AI everywhere, but to use it where the app already knows the relevant form schema, document, file, page, task, and permission context.

What AI Can Do

  • Generate draft form schemas from descriptions.
  • Autofill page or task forms from existing context, files, images, or user instructions.
  • Generate editable document text.
  • Generate page or document images.
  • Summarize uploaded files for review or search.
  • Create draft categories, projects, model tasks, pages, and schedules.
  • Use permission-checked internal search tools when a prompt is configured to allow them.
  • Answer questions using only the records, files, and tasks the current user can access.
  • Suggest workspace-native actions, such as adding pages, tasks, files, forms, or projects in the right place.

Per-User AI Access

An Administrator chooses a separate AI level for each ordinary user; only the primary Owner edits another Administrator’s account. None disables AI features, Ask allows workspace questions and saved answers, and Create also enables generation, autofill, summarization, Organize, Create, and proposal execution. Create includes Ask.

This setting is independent from user groups and permissions, so a user can have broad workspace permissions without AI access, or AI access that still operates only within a narrow set of permitted records.

AI Reports by Email

Registered users can email questions, requests, or attachments to Lagniappe. The app routes each message to Ask, Create, or Organize, saves the result as an AI report, and emails a link back.

  • Ask answers questions using permitted workspace context.
  • Create drafts proposed workspace content from a request.
  • Organize analyzes attached files into a proposed set of workspace changes.

Proposed changes never apply themselves. The recipient must sign in with normal permissions to review and run them. Live inbound addresses are shown only to signed-in users of this installation.

Permission-Scoped Context

AI tools use the same permission boundaries as the rest of Lagniappe. When a user asks a question or starts an AI workflow, the app assembles context through server-side permission checks instead of handing the AI an owner-level view of the workspace.

For example, if someone is invited only to an old house category, they can ask about electricians, repairs, files, or tasks in that category. Other houses, other categories, and records they cannot view are left out of search results and AI tool responses.

Think of AI as another reader acting on behalf of the current user, not as a separate administrator. If the user cannot see a record in Lagniappe, the AI should not receive that record as context.

Privacy Model

AI requests go through your own Google Cloud project’s Vertex AI configuration. That is a meaningfully different posture from pasting private data into a consumer chatbot account, though it is still a cloud AI service: prompts, files, and generated output are sent to Google for processing under your cloud terms and settings.

The training boundary is the important difference. Consumer AI products may use conversations to improve or train their services depending on the product, plan, and account settings. Google Cloud’s managed Gemini model terms say Google will not use customer data to train or fine-tune AI/ML models without prior permission or instruction. That is stronger than relying on a consumer account toggle, but it is not the same as local-only processing.

If zero data retention matters for your installation, choose those settings intentionally. Google’s guidance calls out several retention paths to review, including abuse-monitoring exceptions, request-response logging, cached data, and grounding features that may keep limited logs: Gemini Enterprise Agent Platform zero data retention. Lagniappe setup can help disable Vertex AI caching, but that is one control rather than a blanket guarantee for every AI feature or future provider setting.

Permissions help avoid accidental oversharing into AI calls. As with any AI output, a quick human look at generated text and extracted values is a good habit before relying on them.

Cost

AI usage is pay-per-use through your cloud project. Text operations are usually small. Image generation and large file analysis cost more, and provider pricing can shift over time.

Budget alerts in Google Cloud are an easy safety net. If you turn on search grounding, file processing, image generation, or generate pages in volume, it is worth glancing at usage a little more often.

You can adjust AI settings at any time with:

./setup.sh ai

Good to Know

AI is a helpful collaborator rather than an oracle. A few things to keep in mind when reviewing its output:

  • Generated schemas can look reasonable and still need a tweak or two.
  • Autofill may skip a value, or fill one in that you would rather leave blank.
  • File summaries occasionally miss a detail that matters.
  • Search-grounded answers can lean on the web rather than your own policy or records.
  • Model names, pricing, and retention settings live with the provider and can change.