Basic checks

• I searched existing issues - this hasn't been reported
• I can reproduce this consistently
• This is a RubyLLM bug, not my application code

What's broken?

The convert_schema_to_gemini method in the Gemini provider has several issues that limit its functionality:

1. Description fields are stripped from all schema types - Important schema documentation is lost during transformation
2. Integer and Number types are incorrectly merged - Both integer and number are converted to NUMBER, losing type specificity
3. Inconsistent schema transformation approach - Gemini is the ONLY provider that transforms schemas; all other providers (OpenAI, Anthropic, Mistral, Bedrock) pass schemas directly to their APIs

The gem should either:

• Support all schema fields that Gemini officially supports, or
• Remove schema transformations entirely and let developers handle this themselves (consistent with OpenAI, Anthropic, Mistral, and Bedrock providers)

How to reproduce

Issue 1: Description Fields Stripped From Schema

# Define a schema with a description
schema = {
  type: 'object',
  description: 'An object',
  properties: {
    example: {
      type: "string",
      description: "a brief description about the person's time at the conference"
    }
  },
  required: ['example']
}

chat = RubyLLM.chat(model: 'gemini-2.5-flash')
chat.with_schema(schema)
response = chat.ask('Generate an object')
response.content

{"example" => "hello"} # notice response did not adhere to schema because schema field description was stripped


# Expected: Descriptions should be preserved in API call
# Actual: All description fields are stripped
# Example of expected content: 

{"description" => "The person attended various keynotes and workshops, finding the networking opportunities particularly valuable."}

Issue 2: Integer vs Number Type Loss

# Define schema with both number and integer types
schema = {
  type: 'object',
  properties: {
    number1: {
      type: 'number',
    },
    number2: {
      type: 'integer',
    }
  }
}

chat = RubyLLM.chat(model: 'gemini-2.5-flash')
chat.with_schema(schema)
response = chat.ask('Generate an object')
response.content

{"number1" => 100.5, "number2" => 25.75} # notice number2 is still a number type despite the schema field type labeled as integer

# Expected: price should be NUMBER, quantity should be INTEGER
# Actual: Both are converted to NUMBER type
# Example of expected content: 

{"number1" => 19.99, "number2" => 20}

Expected behavior

All schema fields supported by Gemini should be preserved:

• Objects: description, properties, required
• Strings: description, enum, format, nullable
• Numbers: description, format, minimum, maximum, enum, nullable
• Integers: description, format, minimum, maximum, enum, nullable
• Arrays: description, minItems, maxItems, items, nullable
• Booleans: description, nullable

See steps to reproduce for the expected output from specific examples

What actually happened

The method currently strips important schema information:

• Description fields are completely ignored for all types
• Integer schemas are converted to NUMBER type instead of INTEGER
• Only basic type conversion is performed

Environment

• RubyLLM version: 1.6.2
• Ruby version: 3.4.3
• Provider: Gemini