How to Add AI Features to Your SaaS Product

AI Capabilities You Can Add

The platform provides access to multiple AI providers through simple function calls. Your custom app code can call these functions directly to add AI-powered features:

• Text generation: Generate content, summaries, descriptions, email drafts, or any text output from a prompt. Models range from fast and cheap (GPT-5-nano at 1-2 credits) to powerful reasoning models (GPT-5.2 at 10-15 credits per call).
• Conversational AI: Build a chat interface where users interact with an AI that has context about their data. The platform manages conversation history automatically.
• Knowledge-based answers: Train an AI on your customer's documents using embeddings and RAG, then let it answer questions using that specific knowledge. This is how custom support chatbots and internal knowledge assistants work.
• Data analysis: Send customer data to an AI model and get back summaries, trend analysis, anomaly detection, or actionable recommendations.
• Classification and categorization: Automatically sort incoming data into categories. Classify support tickets by topic, categorize leads by quality, or tag content by type.
• Machine learning predictions: Use the no-code ML system to train classification, regression, or clustering models on customer data and run predictions at zero per-request cost after training.

How AI Calls Work in Your Code

Your custom app functions can call AI models the same way they call the database. The platform provides functions for each AI provider. You pass a system prompt (instructions for the AI), a user message (the actual request), and optionally a model selection. The function returns the AI's response as a string.

For example, a helpdesk SaaS might add an AI feature that suggests a response to a support ticket:

• Customer clicks "Suggest Response" on a ticket
• Your API endpoint reads the ticket content and the account's canned responses from the database
• Your code calls the AI with a system prompt like "You are a support agent. Suggest a response to this ticket using the company's tone and knowledge base."
• The AI returns a draft response that the agent can edit and send

The AI call costs credits based on the model used. GPT-4.1-mini at 2-4 credits per call is good enough for most text generation tasks. Claude Sonnet at 5-8 credits handles complex reasoning. Choose the cheapest model that produces acceptable results for each feature.

AI in the Admin Panel

The admin panel has built-in support for AI assistance in forms. Any field in your database pages can have an AI suggestion button that generates content based on a pre-configured prompt. For example, a product description field can have a button that generates a description from the product name and features.

You can also add full AI chat interfaces to your admin pages using the chatForm field type. This creates a conversational interface where your customers can chat with an AI that has access to their data. See How to Add AI-Built Pages to Your Portal.

Knowledge Base and RAG

If your SaaS needs AI that understands customer-specific information, use the embeddings and RAG (Retrieval Augmented Generation) system. Your customers upload documents, paste text, or point to their website. The platform chunks the content, creates vector embeddings at 3 credits per chunk, and stores them for retrieval.

When the AI needs to answer a question, it searches the embeddings for relevant content, includes that content in the AI prompt, and generates an answer grounded in the customer's actual data. This dramatically reduces hallucinations and ensures answers are specific to each customer's business. See What Is RAG and How Does It Work for the full explanation.

Choosing the Right Model

The platform supports multiple AI models at different price and capability levels:

• GPT-5-nano: Fastest and cheapest. Good for simple classification, short text generation, and data formatting tasks.
• GPT-4.1-mini: Best balance of cost and quality. Good for most text generation, customer-facing responses, and moderate reasoning tasks.
• Claude Sonnet: Strong at nuanced writing and complex analysis. Good for content generation, detailed data analysis, and tasks requiring careful reasoning.
• GPT-5.2 / Claude Opus: Most capable models. Use for complex reasoning, multi-step analysis, and tasks where accuracy is critical and cost is secondary.

For a deeper comparison, see Understanding AI Models: GPT, Claude, and How to Choose.