Skip to main content
When you ask an AI builder to create a CRM or client portal, several systems work together. The model proposes what to do; the agent uses tools to put it into action.

​
The parts behind AI coding

A prompt enters the Zite Agent. Request and Response arrows connect the agent with an AI Model, and a Code changes line connects it to a Sales Team CRM.

The model and agent exchange requests and responses. Shown here with Zite building a sales CRM.

These parts can come from different providers. An AI builder may use another company’s model and provide its own tools and infrastructure.

​
How one request becomes a working feature

For a request like “Add a follow-up date to our sales CRM,” the loop might look like this:
  1. Inspect: The agent sends relevant information to the model and uses tools to look up anything else it needs.
  2. Change: The model proposes changes, and the agent applies them through its tools.
  3. Check: Results or errors go back to the model to guide the next step.
One prompt can trigger several rounds. The agent stops when it finishes, needs your input, or reaches a limit. Try the feature yourself to check that it works for your team.

​
What the model can see

Context is the information supplied with a model request. It can include:
  • Instructions: Your request and the builder’s guidance.
  • Project details: Selected files, database details, and available tools.
  • References: Documents, screenshots, and relevant conversation history.
  • Results: What happened during earlier actions and checks.
The model can process only a limited amount at once. The agent selects or summarizes information, so access to a project does not mean the model sees everything. Missing context can affect the result.
In Zite, you can add context to a workspace, such as a business overview, team policies, or approval steps. The agent can read these shared references without you repeating them in every prompt.
The Context tab with three example references: Business overview, Budget approval policy.txt, and Client access guidelines.

Example workspace context: a business overview in Google Docs, a text file with budget rules, and a note about client access.

​
What runs the finished system?

Once published, your business system relies on:
  • Hosting: Computers that serve its pages and carry out operations.
  • A database: Storage for contacts, projects, orders, and other records.
  • Connected services: Extra capabilities, such as sending emails or summarizing documents with AI.
Everyday actions, such as updating a project status, can run without an AI request. AI-powered features may make their own requests. Some builders provide this infrastructure together; others require separate setup.

​
Why building and running have costs

Building uses model processing and tools. Longer tasks can take more rounds, and failed or discarded attempts can still cost money. Running the finished system uses hosting, data storage, and any connected services. How these costs appear in a subscription varies by builder. For Zite, see How credits work and App Usage.

Related topics

What can I build?Model selectionUse credits efficiently
Last modified on October 22, 2026