Native AI: Mastering 'Generate Response from Model' in FileMaker 2025

The integration of Large Language Models (LLMs) into custom applications has evolved from simple text generation to intelligent agents capable of performing tasks. For years, FileMaker developers relied on brittle Insert from URL calls to achieve this.

With Claris FileMaker 2025, the platform has introduced the Generate Response from Model script step. This is not just a wrapper for an API call; it is a fully integrated Reasoning Engine capable of executing FileMaker logic.

Here is the architectural breakdown of how to use this feature effectively, specifically focusing on the new Agentic Mode.

The Architectural Shift: Agents vs. Chatbots

In previous iterations (and other platforms), AI implementation was often "Stateless Request/Response." You sent a prompt, got text back, and parsed it.

FileMaker 2025 changes this paradigm with Agentic Mode. When enabled, the script step does not just generate text; it analyzes the user's intent and determines if it needs to act before answering. It can run SQL queries, call your Custom Functions, or retrieve images—all within a single script step execution loop.

Prerequisites: The Configuration

Before utilizing the script step, the session must be authenticated. This is handled via Configure AI Account.

# Best Practice: Run this in your OnFirstWindowOpen script
Set Variable [ $apiKey ; Value: Get ( EnvironmentVariable_API_KEY ) ]

# Configure the provider (OpenAI, Anthropic, or Cohere)
Configure AI Account [
    Provider: "OpenAI" ;
    Model: "gpt-4o" ;
    Key: $apiKey
]

Implementation: The 'Generate Response from Model' Step

The power of this step lies in its parameters. It handles the "Reasoning Loop" (Thought -> Tool Call -> Observation -> Answer) internally, saving developers from writing complex recursive scripts.

1. The Context (Message History)

One of the most tedious parts of AI development is managing the "context window." FileMaker 2025 handles this natively via the Messages and Save Message History To options.

Architectural Tip: Use a global variable (e.g., $$chat_context) to persist the conversation across the user's session automatically.

2. Agentic Mode & Tool Definitions

This is where the magic happens. By enabling Agentic Mode, you give the model permission to use specific tools. FileMaker provides two native tools out of the box:

• execute_sql: Allows the LLM to query your data schema directly.
• retrieve_image: Allows semantic search against visual data.

You can also define your own tools using Custom Functions.

Use Case: "Talk to Your Data" (No SQL Knowledge Required)

Imagine a dashboard where an executive asks, "How many units of the 'Red Widget' did we sell in Texas last month?"

Script: AI Data Analyst

# 1. Capture User Input
Set Variable [ $userPrompt ; Value: Dashboard::GlobalInput ]

# 2. Call the Model with Agentic Mode
Generate Response from Model [
    Account Name: "OpenAI" ;
    Model: "gpt-4o" ;
    User Prompt: $userPrompt ;

    # ENABLE AGENTIC MODE
    Agentic: On ;

    # Define Tools (JSON format)
    # We enable the native SQL tool so the model can look up the data
    Tool Definitions: JSONSetElement ( "{}" ;
        [ "tools[0].name" ; "execute_sql" ; JSONString ];
        [ "tools[0].description" ; "Run SQL queries to answer user questions about sales" ; JSONString ]
    ) ;

    # History Management
    Messages: $$chat_context ;
    Save Message History To: $$chat_context ;

    # STREAMING TARGET (Must be a field for real-time visual)
    Response: Dashboard::ChatOutput
]

# 3. Error Handling
If [ Get ( LastError ) ≠ 0 ]
    Perform Script [ "Log AI Failure" ]
End If

How It Works Under the Hood

1. Intent Analysis: The model sees the question about "sales in Texas."
2. Tool Selection: It recognizes it has the execute_sql tool.
3. Execution: It generates the SQL (SELECT SUM(Quantity) FROM Sales WHERE State = 'TX'...), runs it against your FileMaker file, and gets the result.
4. Synthesis: It formulates the final answer: "We sold 450 units of the Red Widget in Texas last month."

All of this happens inside the single Generate Response from Model step.

Trade-offs and Best Practices

Latency vs. Streaming

If you target a Field, FileMaker 2025 streams the text response in real-time, providing an excellent UX. If you target a Variable, the script waits until the entire generation is complete.