Skip to main content

​
Overview

When you ask Zite to build something, you are working with Zite’s AI agent. Zite uses AI models (LLMs) to power that agent, then gives it the context and tools it needs to create and update your project. The agent can create the parts of a connected business system, including:
  • Zite apps and pages
  • Shared databases, tables, fields, and relationships
  • Forms for collecting or updating information
  • Workflows, permissions, and integrations
This is what people often mean by AI coding: describing the outcome you want in everyday language and letting an agent translate that request into working app structure, logic, and design.
An AI model exchanging requests and responses with the Zite agent, which builds a Sales Team CRM app
Zite is not a direct chat window with an LLM. Zite combines LLMs with Zite’s app-building tools, product rules, and your project’s context so the agent can make changes inside Zite.

​
What happens when you send a request?

The exact process depends on your request, but an AI build usually follows this pattern:
1

Understand the goal

The agent interprets what you want to achieve, who the app is for, and any constraints you included. A request such as “create a client onboarding portal” gives the agent a goal; details about the users, records, and workflow make the goal clearer.
2

Gather relevant context

The agent looks at the parts of your project that matter to the request, such as existing pages, app logic, workflows, database tables, fields, permissions, and attached references. It uses this context to understand what already exists before deciding what to change.
3

Plan the changes

The agent works out which app elements, data structures, and workflow steps are needed. For larger changes, use Plan mode to review the proposed direction before the app is built.
4

Use Zite's building tools

The agent creates or updates the relevant pages, components, forms, tables, fields, workflows, and permissions using Zite’s tools. It is working with the structure of your Zite project rather than only producing a block of code in a chat response.
5

Report the result

Zite summarizes the work so you can review and test it. AI can make mistakes, so check the result against your requirements, especially when the request changes data, permissions, or business-critical workflows.

​
What is context?

Context is the information the agent can use to understand your request. It can include:
  • Your current prompt
  • Relevant messages and decisions from the chat
  • Zite’s system prompt, which defines how the agent should work within Zite
  • The structure of your app, including pages, components, workflows, and generated logic
  • Your database tables, fields, relationships, and permissions
  • Relevant integrations and the actions available through them
  • Files, screenshots, mockups, or other references you attach
  • The mode and model you selected
Context helps the agent make a change that fits your existing project. For example, “add a status filter” is easier to implement when the agent can see which page contains the table, which database field stores the status, and which visual patterns the app already uses. More context is not always better. A long chat can contain old decisions, previous experiments, or instructions that no longer apply. Keep requests focused, start a new chat for an unrelated feature, and say what the agent should preserve or leave unchanged.
A useful prompt explains the outcome, where it should happen, the data or workflow involved, and any constraints. Clear context gives the agent less to guess and usually means less rework.

​
Why AI work uses credits

When a credit-consuming Zite AI feature runs, Zite sends information to an LLM and receives work back. Zite incurs a cost from the LLM provider for that AI processing, including the work needed to understand your request, inspect context, reason about a solution, and produce actions. That cost is incurred even when the final result is not what you expected. For example, an AI request can use credits if:
  • The app has an error after the changes are applied
  • The agent misunderstands part of your request
  • The build stops partway through
  • You decide to undo the changes
  • The result needs another request to correct it
Credits cover the AI work that took place, not only the changes that ended up working perfectly. This is similar to paying for a developer’s time spent investigating and attempting a change: an unsuccessful attempt still required real work.
An error in the app does not mean that no AI service was used. If an AI request ran, it may have consumed credits even if the request produced an error or the result needs to be fixed.
The Fix it for me repair action is an exception: it is free and does not consume credits. See How credits work for the current list of actions that use credits and actions that do not. The amount of credits used depends on the work involved. Larger apps, longer context, more capable models, and requests that affect multiple screens, tables, workflows, or integrations generally require more AI processing.

Related topics

What can I build?Agent ModesModel SelectionHow credits workUse credits efficiently
Last modified on August 3, 2026