Skip to content

Latest commit

 

History

History
77 lines (58 loc) · 3.14 KB

README.md

File metadata and controls

77 lines (58 loc) · 3.14 KB

Swift Function Tools Experiment

This project demonstrates how to integrate swift-json-schema into a Swift-based conversational assistant. The assistant, powered by the OpenAI API using the MacPaw Swift wrapper, can invoke functions (tools) that expect strictly defined, schema-validated arguments—eliminating guesswork and brittle JSON parsing.

By defining schemas for each tool, you ensure that when the model requests a function call, its arguments match the precise format and types you expect. This leads to more reliable interactions, better error handling, and easier integration with external services (like weather APIs or order tracking systems).

Defining a Schema:

let weatherSchema = JSONObject {
  JSONProperty(key: "location") {
    JSONString()
      .description("The city")
  }
  .required()
}

Registering a Tool:

service.registerTool(name: "get_weather", schema: weatherSchema, handler: weather(location:))

Here, get_weather expects a location string. When the model calls it, the arguments are automatically validated. The handler receives a strongly typed Swift String—no manual JSON parsing needed.

Another Tool With a Struct Schema:

@Schemable
struct DeliveryLookupData {
  @SchemaOptions(description: "The customer's order ID.")
  let orderID: String
}

service.registerTool(
  name: "get_delivery_date",
  description: "Get the delivery date for a customer's order.",
  schema: DeliveryLookupData.schema,
  handler: deliveryDate(_:)
)

In this example, the schema is derived from a Swift struct decorated with @Schemable. When the model calls get_delivery_date, the orderID is extracted and validated automatically.

See the full code here.

Example Output

➜  SwiftFunctionToolsExperiment git:(main) swift run SwiftFunctionToolsExperiment "How much does shipping cost for express, 6.2 kg, and priority? What is the weather in Paris? And do you know that status of order 65445436"
Building for debugging...
[19/19] Applying SwiftFunctionToolsExperiment
Build of product 'SwiftFunctionToolsExperiment' complete! (5.92s)
Hello from calculateShippingCost
Mocking call for ShippingEstimateRequest(weight: 6.2, priority: true, deliveryType: SwiftFunctionToolsExperiment.DeliveryType.express, extras: nil)
Hello from weather
Mocking call for Paris
Hello from deliveryDate
Mocking call for 65445436
Final result: Here's the information you requested:
- **Shipping Cost** for express delivery, 6.2 kg package, and priority: **$20**
- **Weather in Paris**: It's currently **32°C**
- **Status of Order 65445436**: Scheduled delivery is on **15th January 2021**.

Run locally

To run the Swift executable locally, follow these steps:

  1. Checkout the repository

  2. In the root directory of the project, create a file named .env and add your OpenAI API key:

OPENAI_KEY="<your-api-key>"

  1. Run the Swift executable

swift run SwiftFunctionToolsExperiment "How much does shipping cost for express, 6.2 kg, and priority?"