ai-lite 0.3.0 → 0.5.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: cd1b2ead0a8ae48d9617b9e73460649f817fc158430e11cc7b9ac798a86846d2
-  data.tar.gz: f3dad03f56f3907e537dc8c72d63bf7e57e069d4b28ef666151d348b98f6b3e6
+  metadata.gz: cb52e45732e37e363b41144d47cc4136dab1feb450890f10cf84f7cc15593b2a
+  data.tar.gz: 2a05669e6fccdf8da0d1a3a68dc21598806e6072de2fda487e9b436751add386
 SHA512:
-  metadata.gz: 6e70409efd1d697b2578ace5b878e74ef4e6214efbcbae2c295aae066e0e9ce4887cd39c95fcdde8bc73e0f6b0cfe60434dc7f1ca2611c77e649cc645b94351a
-  data.tar.gz: 1faba64c32c14038f18c1d50a2f73bd92a0c940d8b303b3eecb89c2193d944d3b189334c5ecde81f55e8268db01e01e562686b74f20111105f0c0fbbeea387c6
+  metadata.gz: 307ffdbea749aac0eeca74b678d8af55a96add4924e1ef5e5f9157a83138b0929adc2db3282d79a924a24929d8c532fe1742f8ecb801409cf4d5c38fff3639ee
+  data.tar.gz: a50f9ef6171f3ee7249fc46d02e77698820272df70fe3ddbef777f6ab3159879c92c6d6d5fd7eddf29ceda3387308480a2622f95772b85b21fdbc2037808eda7

data/README.md

@@ -19,6 +19,8 @@ It is not meant to replace the official OpenAI SDK. It is a small wrapper for pr
 ai.chat("Say hello")
 ai.moderate("User submitted text")
 ai.embed("Text to vectorize")
+ai.image("A simple app icon")
+ai.speak("Read this aloud")
 ```
 
 ## Usage
@@ -51,6 +53,9 @@ AiLite.configure do |config|
   config.model = "gpt-5.5"
   config.moderation_model = "omni-moderation-latest"
   config.embedding_model = "text-embedding-3-small"
+  config.image_model = "gpt-image-2"
+  config.speech_model = "gpt-4o-mini-tts"
+  config.speech_voice = "alloy"
   config.timeout = 120
   config.max_output_tokens = 2000
 end
@@ -95,6 +100,23 @@ The default model is `gpt-5.5`.
 
 The OpenAI API URL is fixed to `https://api.openai.com/v1/responses`.
 
+### Multi-Turn Chat
+
+Responses include a `response_id` that can be passed back through `options` as `previous_response_id`:
+
+```ruby
+first = ai.chat("Tell me a short joke.")
+
+follow_up = ai.chat(
+  "Explain why that is funny.",
+  options: {
+    previous_response_id: first["response_id"]
+  }
+)
+
+puts follow_up["content"]
+```
+
 ## Moderation
 
 Use `moderate` to classify user-submitted text or images for potentially harmful content before saving, publishing, or sending it into another AI call.
@@ -239,21 +261,156 @@ result["content"]       # embedding vector
 result["raw"]["usage"]  # token usage
 ```
 
-## Multi-Turn Chat
+## Images
 
-Responses include a `response_id` that can be passed back through `options` as `previous_response_id`:
+Use `image` to generate an image from a prompt.
 
 ```ruby
-first = ai.chat("Tell me a short joke.")
+result = ai.image("A clean Ruby gem logo on a white background")
+image_data = result["content"]
+```
 
-follow_up = ai.chat(
-  "Explain why that is funny.",
-  options: {
-    previous_response_id: first["response_id"]
-  }
+By default, `content` is the base64-encoded generated image:
+
+```ruby
+{
+  "content" => "iVBORw0KGgo...",
+  "response_id" => nil,
+  "status" => 200,
+  "error" => nil,
+  "raw" => nil
+}
+```
+
+Write the generated image bytes directly to a file with `output_path`:
+
+```ruby
+result = ai.image(
+  "A clean Ruby gem logo on a white background",
+  output_path: "tmp/logo.png"
 )
+```
 
-puts follow_up["content"]
+`image` sends a `POST` request to `/v1/images/generations` with:
+
+- `model`
+- `prompt`
+- optional `size`
+- optional `quality`
+- optional `background`
+- optional `output_format`
+- optional `debug`
+- optional extra `options`
+
+The default image model is `gpt-image-2`.
+
+Use `output_format` to request `png`, `webp`, or `jpeg` output:
+
+```ruby
+result = ai.image(
+  "A transparent app icon",
+  background: "transparent",
+  output_format: "webp",
+  output_path: "tmp/icon.webp"
+)
+```
+
+Pass `debug: true` to include the raw OpenAI response, including usage when returned:
+
+```ruby
+result = ai.image("A tiny robot sticker", debug: true)
+
+result["content"]       # base64 image data
+result["raw"]["usage"]  # token usage, when returned
+```
+
+## Speech
+
+Use `speak` to generate audio from text.
+
+```ruby
+result = ai.speak("Hello from AI Lite")
+audio_bytes = result["content"]
+```
+
+By default, `content` is the raw audio bytes returned by OpenAI:
+
+```ruby
+{
+  "content" => "...binary audio bytes...",
+  "response_id" => nil,
+  "status" => 200,
+  "error" => nil,
+  "raw" => nil
+}
+```
+
+Write the generated audio directly to a file with `output_path`:
+
+```ruby
+result = ai.speak(
+  "Hello from AI Lite",
+  output_path: "tmp/hello.mp3"
+)
+```
+
+When `output_path` is used, `content` is file metadata:
+
+```ruby
+{
+  "content" => {
+    "path" => "tmp/hello.mp3",
+    "bytes" => 12345,
+    "format" => "mp3"
+  },
+  "response_id" => nil,
+  "status" => 200,
+  "error" => nil,
+  "raw" => nil
+}
+```
+
+Use `base64: true` when you want text-safe audio data that can be transported in JSON and decoded later:
+
+```ruby
+result = ai.speak("Hello from AI Lite", base64: true)
+
+File.binwrite("tmp/hello.mp3", Base64.decode64(result["content"]))
+```
+
+`speak` sends a `POST` request to `/v1/audio/speech` with:
+
+- `model`
+- `input`
+- `voice`
+- optional `response_format`
+- optional `speed`
+- optional `instructions`
+- optional `debug`
+- optional extra `options`
+
+The default speech model is `gpt-4o-mini-tts`.
+The default speech voice is `alloy`.
+The default response format is `mp3`.
+
+Set `voice` per call when you want a different built-in voice:
+
+```ruby
+result = ai.speak(
+  "Hello from AI Lite",
+  voice: "sage",
+  output_path: "tmp/hello.mp3"
+)
+```
+
+Use `response_format` to request `mp3`, `opus`, `aac`, `flac`, `wav`, or `pcm` output:
+
+```ruby
+result = ai.speak(
+  "Export this as a WAV file",
+  response_format: "wav",
+  output_path: "tmp/hello.wav"
+)
 ```
 
 ## Return Shape

data/lib/ai_lite/version.rb

@@ -1,3 +1,3 @@
 class AiLite
-  VERSION = "0.3.0".freeze
+  VERSION = "0.5.0".freeze
 end

data/lib/ai_lite.rb

@@ -9,6 +9,10 @@ class AiLite
   DEFAULT_MODEL = "gpt-5.5".freeze
   DEFAULT_MODERATION_MODEL = "omni-moderation-latest".freeze
   DEFAULT_EMBEDDING_MODEL = "text-embedding-3-small".freeze
+  DEFAULT_IMAGE_MODEL = "gpt-image-2".freeze
+  DEFAULT_SPEECH_MODEL = "gpt-4o-mini-tts".freeze
+  DEFAULT_SPEECH_VOICE = "alloy".freeze
+  DEFAULT_SPEECH_FORMAT = "mp3".freeze
   DEFAULT_TIMEOUT = 120
   DEFAULT_MAX_OUTPUT_TOKENS = 2000
   IMAGE_MIME_TYPES = {
@@ -20,13 +24,16 @@ class AiLite
   }.freeze
 
   class Configuration
-    attr_accessor :api_key, :model, :moderation_model, :embedding_model, :timeout, :max_output_tokens
+    attr_accessor :api_key, :model, :moderation_model, :embedding_model, :image_model, :speech_model, :speech_voice, :timeout, :max_output_tokens
 
     def initialize
       @api_key = nil
       @model = DEFAULT_MODEL
       @moderation_model = DEFAULT_MODERATION_MODEL
       @embedding_model = DEFAULT_EMBEDDING_MODEL
+      @image_model = DEFAULT_IMAGE_MODEL
+      @speech_model = DEFAULT_SPEECH_MODEL
+      @speech_voice = DEFAULT_SPEECH_VOICE
       @timeout = DEFAULT_TIMEOUT
       @max_output_tokens = DEFAULT_MAX_OUTPUT_TOKENS
     end
@@ -65,20 +72,31 @@ class AiLite
       client.embed(input, **kwargs)
     end
 
+    def image(prompt, **kwargs)
+      client.image(prompt, **kwargs)
+    end
+
+    def speak(text, **kwargs)
+      client.speak(text, **kwargs)
+    end
+
     def reset_client!
       @client = nil
     end
   end
 
-  attr_reader :api_key, :model, :moderation_model, :embedding_model, :timeout, :max_output_tokens, :headers
+  attr_reader :api_key, :model, :moderation_model, :embedding_model, :image_model, :speech_model, :speech_voice, :timeout, :max_output_tokens, :headers
 
-  def initialize(api_key: nil, model: nil, moderation_model: nil, embedding_model: nil, timeout: nil, max_output_tokens: nil)
+  def initialize(api_key: nil, model: nil, moderation_model: nil, embedding_model: nil, image_model: nil, speech_model: nil, speech_voice: nil, timeout: nil, max_output_tokens: nil)
     @api_key = api_key || self.class.configuration.api_key || ENV["OPENAI_API_KEY"] || ENV["OPEN_AI_TOKEN"]
     raise ArgumentError, "Missing OpenAI API key" if @api_key.to_s.strip.empty?
 
     @model = model || self.class.configuration.model
     @moderation_model = moderation_model || self.class.configuration.moderation_model
     @embedding_model = embedding_model || self.class.configuration.embedding_model
+    @image_model = image_model || self.class.configuration.image_model
+    @speech_model = speech_model || self.class.configuration.speech_model
+    @speech_voice = speech_voice || self.class.configuration.speech_voice
     @timeout = timeout || self.class.configuration.timeout
     @max_output_tokens = max_output_tokens || self.class.configuration.max_output_tokens
     @headers = {
@@ -124,6 +142,42 @@ class AiLite
     prettify_data(status: "unknown", error: e.message, raw: nil, debug: debug)
   end
 
+  def image(prompt, model: nil, size: nil, quality: nil, background: nil, output_format: nil, output_path: nil, debug: false, options: {})
+    payload = options.merge(
+      model: model || image_model,
+      prompt: prompt
+    )
+    payload[:size] = size if size
+    payload[:quality] = quality if quality
+    payload[:background] = background if background
+    payload[:output_format] = output_format if output_format
+
+    extract_image(post(payload, endpoint: image_endpoint), output_path: output_path, debug: debug)
+  rescue => e
+    prettify_data(status: "unknown", error: e.message, raw: nil, debug: debug)
+  end
+
+  def speak(text, model: nil, voice: nil, response_format: nil, speed: nil, instructions: nil, output_path: nil, base64: false, debug: false, options: {})
+    payload = options.merge(
+      model: model || speech_model,
+      input: text,
+      voice: voice || speech_voice
+    )
+    payload[:response_format] = response_format if response_format
+    payload[:speed] = speed if speed
+    payload[:instructions] = instructions if instructions
+
+    extract_speech(
+      post(payload, endpoint: speech_endpoint),
+      output_path: output_path,
+      base64: base64,
+      response_format: payload[:response_format] || payload["response_format"] || DEFAULT_SPEECH_FORMAT,
+      debug: debug
+    )
+  rescue => e
+    prettify_data(status: "unknown", error: e.message, raw: nil, debug: debug)
+  end
+
   private
 
   def post(payload, endpoint: response_endpoint)
@@ -155,6 +209,14 @@ class AiLite
     "#{API_BASE_URL}/embeddings"
   end
 
+  def image_endpoint
+    "#{API_BASE_URL}/images/generations"
+  end
+
+  def speech_endpoint
+    "#{API_BASE_URL}/audio/speech"
+  end
+
   def extract_content(response, debug: false)
     status = response.code.to_i
     parsed_response = JSON.parse(response.body)
@@ -238,6 +300,65 @@ class AiLite
     prettify_data(status: response_status(response), error: e.message, raw: nil, debug: debug)
   end
 
+  def extract_image(response, output_path:, debug: false)
+    status = response.code.to_i
+    parsed_response = JSON.parse(response.body)
+
+    unless success_status?(status)
+      return prettify_data(
+        status: status,
+        error: error_message(parsed_response),
+        response_id: parsed_response["id"],
+        raw: parsed_response,
+        debug: debug
+      )
+    end
+
+    content = image_content(parsed_response)
+    write_image_output(output_path, content) if output_path
+
+    prettify_data(
+      status: status,
+      content: content,
+      response_id: parsed_response["id"],
+      raw: parsed_response,
+      debug: debug
+    )
+  rescue JSON::ParserError => e
+    prettify_data(status: response_status(response), error: e.message, raw: response&.body, debug: debug)
+  rescue => e
+    prettify_data(status: response_status(response), error: e.message, raw: nil, debug: debug)
+  end
+
+  def extract_speech(response, output_path:, base64:, response_format:, debug: false)
+    status = response.code.to_i
+
+    unless success_status?(status)
+      parsed_response = parse_error_response(response.body)
+
+      return prettify_data(
+        status: status,
+        error: error_message(parsed_response),
+        response_id: parsed_response.is_a?(Hash) ? parsed_response["id"] : nil,
+        raw: parsed_response,
+        debug: debug
+      )
+    end
+
+    audio = response.body
+    File.binwrite(output_path, audio) if output_path
+
+    prettify_data(
+      status: status,
+      content: speech_content(audio, output_path: output_path, base64: base64, response_format: response_format),
+      response_id: nil,
+      raw: audio,
+      debug: debug
+    )
+  rescue => e
+    prettify_data(status: response_status(response), error: e.message, raw: nil, debug: debug)
+  end
+
   def extract_output_text(raw)
     Array(raw["output"]).flat_map do |item|
       next [] unless item.is_a?(Hash) && item["type"] == "message"
@@ -312,6 +433,34 @@ class AiLite
     multiple ? embeddings : embeddings.first
   end
 
+  def image_content(raw)
+    image = Array(raw["data"]).find { |item| item.is_a?(Hash) && item["b64_json"] }
+    image && image["b64_json"]
+  end
+
+  def write_image_output(path, content)
+    raise "No image data returned" if content.to_s.empty?
+
+    File.binwrite(path, Base64.decode64(content))
+  end
+
+  def speech_content(audio, output_path:, base64:, response_format:)
+    return Base64.strict_encode64(audio) if base64
+    return audio unless output_path
+
+    {
+      "path" => output_path,
+      "bytes" => audio.bytesize,
+      "format" => response_format
+    }
+  end
+
+  def parse_error_response(body)
+    JSON.parse(body)
+  rescue JSON::ParserError
+    body
+  end
+
   def success_status?(status)
     status >= 200 && status < 300
   end

data/test/ai_lite_test.rb

@@ -33,6 +33,9 @@ class AiLiteTest < Minitest::Test
       client = AiLite.new(
         api_key: "explicit-key",
         model: "gpt-test",
+        image_model: "gpt-image-test",
+        speech_model: "gpt-speech-test",
+        speech_voice: "verse",
         timeout: 10
       )
 
@@ -40,6 +43,9 @@ class AiLiteTest < Minitest::Test
       assert_equal "gpt-test", client.model
       assert_equal "omni-moderation-latest", client.moderation_model
       assert_equal "text-embedding-3-small", client.embedding_model
+      assert_equal "gpt-image-test", client.image_model
+      assert_equal "gpt-speech-test", client.speech_model
+      assert_equal "verse", client.speech_voice
       assert_equal 10, client.timeout
       assert_equal 2000, client.max_output_tokens
       assert_equal "Bearer explicit-key", client.headers["Authorization"]
@@ -54,6 +60,9 @@ class AiLiteTest < Minitest::Test
         config.model = "gpt-config"
         config.moderation_model = "omni-moderation-test"
         config.embedding_model = "text-embedding-test"
+        config.image_model = "gpt-image-test"
+        config.speech_model = "gpt-speech-test"
+        config.speech_voice = "verse"
         config.timeout = 15
         config.max_output_tokens = 750
       end
@@ -64,6 +73,9 @@ class AiLiteTest < Minitest::Test
       assert_equal "gpt-config", client.model
       assert_equal "omni-moderation-test", client.moderation_model
       assert_equal "text-embedding-test", client.embedding_model
+      assert_equal "gpt-image-test", client.image_model
+      assert_equal "gpt-speech-test", client.speech_model
+      assert_equal "verse", client.speech_voice
       assert_equal 15, client.timeout
       assert_equal 750, client.max_output_tokens
       assert_same client, AiLite.client
@@ -87,6 +99,9 @@ class AiLiteTest < Minitest::Test
       config.model = "gpt-config"
       config.moderation_model = "omni-moderation-config"
       config.embedding_model = "text-embedding-config"
+      config.image_model = "gpt-image-config"
+      config.speech_model = "gpt-speech-config"
+      config.speech_voice = "sage"
       config.timeout = 15
       config.max_output_tokens = 750
     end
@@ -96,6 +111,9 @@ class AiLiteTest < Minitest::Test
       model: "gpt-explicit",
       moderation_model: "omni-moderation-explicit",
       embedding_model: "text-embedding-explicit",
+      image_model: "gpt-image-explicit",
+      speech_model: "gpt-speech-explicit",
+      speech_voice: "coral",
       timeout: 5,
       max_output_tokens: 300
     )
@@ -104,6 +122,9 @@ class AiLiteTest < Minitest::Test
     assert_equal "gpt-explicit", client.model
     assert_equal "omni-moderation-explicit", client.moderation_model
     assert_equal "text-embedding-explicit", client.embedding_model
+    assert_equal "gpt-image-explicit", client.image_model
+    assert_equal "gpt-speech-explicit", client.speech_model
+    assert_equal "coral", client.speech_voice
     assert_equal 5, client.timeout
     assert_equal 300, client.max_output_tokens
   end
@@ -525,6 +546,245 @@ class AiLiteTest < Minitest::Test
     end
   end
 
+  def test_image_sends_post_to_images_with_default_payload
+    client = AiLite.new(api_key: "token-abc")
+    image_data = Base64.strict_encode64("fake image")
+
+    with_stubbed_http(image_response(b64_json: image_data)) do |captured, _response|
+      result = client.image("A small ruby gem logo")
+      request = captured[:http].last_request
+      payload = JSON.parse(request.body)
+
+      assert_equal image_data, result["content"]
+      assert_nil result["response_id"]
+      assert_equal 200, result["status"]
+      assert_nil result["error"]
+      assert_nil result["raw"]
+      assert_equal "api.openai.com", captured[:host]
+      assert_equal 443, captured[:port]
+      assert_equal true, captured[:use_ssl]
+      assert_instance_of Net::HTTP::Post, request
+      assert_equal "/v1/images/generations", request.path
+      assert_equal "Bearer token-abc", request["Authorization"]
+      assert_equal "application/json", request["Content-Type"]
+      assert_equal "gpt-image-2", payload["model"]
+      assert_equal "A small ruby gem logo", payload["prompt"]
+    end
+  end
+
+  def test_image_includes_options_size_quality_background_output_format_and_model
+    client = AiLite.new(api_key: "token-abc")
+
+    with_stubbed_http(image_response) do |captured, _response|
+      client.image(
+        "A transparent app icon",
+        model: "gpt-image-test",
+        size: "1024x1536",
+        quality: "high",
+        background: "transparent",
+        output_format: "webp",
+        options: {
+          moderation: "auto",
+          output_compression: 80
+        }
+      )
+      payload = JSON.parse(captured[:http].last_request.body)
+
+      assert_equal "gpt-image-test", payload["model"]
+      assert_equal "A transparent app icon", payload["prompt"]
+      assert_equal "1024x1536", payload["size"]
+      assert_equal "high", payload["quality"]
+      assert_equal "transparent", payload["background"]
+      assert_equal "webp", payload["output_format"]
+      assert_equal "auto", payload["moderation"]
+      assert_equal 80, payload["output_compression"]
+    end
+  end
+
+  def test_image_uses_class_level_configured_client
+    AiLite.configure do |config|
+      config.api_key = "configured-key"
+      config.image_model = "gpt-image-config"
+    end
+
+    with_stubbed_http(image_response) do |captured, _response|
+      AiLite.image("Use configured defaults")
+      payload = JSON.parse(captured[:http].last_request.body)
+
+      assert_equal "gpt-image-config", payload["model"]
+      assert_equal "Bearer configured-key", captured[:http].last_request["Authorization"]
+    end
+  end
+
+  def test_image_writes_decoded_content_to_output_path
+    client = AiLite.new(api_key: "token-abc")
+    image_data = Base64.strict_encode64("fake image")
+
+    Tempfile.create(["generated", ".png"]) do |file|
+      with_stubbed_http(image_response(b64_json: image_data)) do |_captured, _response|
+        result = client.image("A file output", output_path: file.path)
+
+        assert_equal image_data, result["content"]
+        assert_equal "fake image", File.binread(file.path)
+      end
+    end
+  end
+
+  def test_image_debug_true_returns_raw_usage
+    client = AiLite.new(api_key: "token-abc")
+    image_data = Base64.strict_encode64("fake image")
+
+    with_stubbed_http(image_response(b64_json: image_data)) do |_captured, _response|
+      result = client.image("A debuggable image", debug: true)
+
+      assert_equal image_data, result["content"]
+      assert_equal({ "input_tokens" => 10, "output_tokens" => 20, "total_tokens" => 30 }, result["raw"]["usage"])
+    end
+  end
+
+  def test_image_output_path_without_image_data_returns_standard_envelope
+    client = AiLite.new(api_key: "token-abc")
+
+    Tempfile.create(["generated", ".png"]) do |file|
+      with_stubbed_http(image_response(b64_json: nil)) do |_captured, _response|
+        result = client.image("A missing image", output_path: file.path)
+
+        assert_nil result["content"]
+        assert_nil result["response_id"]
+        assert_equal 200, result["status"]
+        assert_equal "No image data returned", result["error"]
+        assert_nil result["raw"]
+      end
+    end
+  end
+
+  def test_speak_sends_post_to_audio_speech_with_default_payload
+    client = AiLite.new(api_key: "token-abc")
+
+    with_stubbed_http(speech_response("fake audio")) do |captured, _response|
+      result = client.speak("Read this aloud")
+      request = captured[:http].last_request
+      payload = JSON.parse(request.body)
+
+      assert_equal "fake audio", result["content"]
+      assert_nil result["response_id"]
+      assert_equal 200, result["status"]
+      assert_nil result["error"]
+      assert_nil result["raw"]
+      assert_equal "api.openai.com", captured[:host]
+      assert_equal 443, captured[:port]
+      assert_equal true, captured[:use_ssl]
+      assert_instance_of Net::HTTP::Post, request
+      assert_equal "/v1/audio/speech", request.path
+      assert_equal "Bearer token-abc", request["Authorization"]
+      assert_equal "application/json", request["Content-Type"]
+      assert_equal "gpt-4o-mini-tts", payload["model"]
+      assert_equal "Read this aloud", payload["input"]
+      assert_equal "alloy", payload["voice"]
+    end
+  end
+
+  def test_speak_includes_options_response_format_speed_instructions_model_and_voice
+    client = AiLite.new(api_key: "token-abc")
+
+    with_stubbed_http(speech_response) do |captured, _response|
+      client.speak(
+        "Use a clear support tone",
+        model: "gpt-speech-test",
+        voice: "sage",
+        response_format: "wav",
+        speed: 1.2,
+        instructions: "Speak warmly.",
+        options: {
+          stream_format: "audio"
+        }
+      )
+      payload = JSON.parse(captured[:http].last_request.body)
+
+      assert_equal "gpt-speech-test", payload["model"]
+      assert_equal "Use a clear support tone", payload["input"]
+      assert_equal "sage", payload["voice"]
+      assert_equal "wav", payload["response_format"]
+      assert_equal 1.2, payload["speed"]
+      assert_equal "Speak warmly.", payload["instructions"]
+      assert_equal "audio", payload["stream_format"]
+    end
+  end
+
+  def test_speak_uses_class_level_configured_client
+    AiLite.configure do |config|
+      config.api_key = "configured-key"
+      config.speech_model = "gpt-speech-config"
+      config.speech_voice = "marin"
+    end
+
+    with_stubbed_http(speech_response) do |captured, _response|
+      AiLite.speak("Use configured defaults")
+      payload = JSON.parse(captured[:http].last_request.body)
+
+      assert_equal "gpt-speech-config", payload["model"]
+      assert_equal "marin", payload["voice"]
+      assert_equal "Bearer configured-key", captured[:http].last_request["Authorization"]
+    end
+  end
+
+  def test_speak_writes_audio_to_output_path_and_returns_metadata
+    client = AiLite.new(api_key: "token-abc")
+
+    Tempfile.create(["speech", ".mp3"]) do |file|
+      with_stubbed_http(speech_response("fake audio")) do |_captured, _response|
+        result = client.speak("Save this", output_path: file.path)
+
+        assert_equal(
+          {
+            "path" => file.path,
+            "bytes" => 10,
+            "format" => "mp3"
+          },
+          result["content"]
+        )
+        assert_equal "fake audio", File.binread(file.path)
+      end
+    end
+  end
+
+  def test_speak_base64_true_returns_encoded_audio
+    client = AiLite.new(api_key: "token-abc")
+
+    with_stubbed_http(speech_response("fake audio")) do |_captured, _response|
+      result = client.speak("Encode this", base64: true)
+
+      assert_equal Base64.strict_encode64("fake audio"), result["content"]
+      assert_nil result["raw"]
+    end
+  end
+
+  def test_speak_debug_true_returns_raw_audio
+    client = AiLite.new(api_key: "token-abc")
+
+    with_stubbed_http(speech_response("fake audio")) do |_captured, _response|
+      result = client.speak("Debug this", debug: true)
+
+      assert_equal "fake audio", result["content"]
+      assert_equal "fake audio", result["raw"]
+    end
+  end
+
+  def test_speak_http_errors_return_standard_envelope
+    client = AiLite.new(api_key: "token-abc")
+    body = JSON.generate("error" => { "message" => "Unsupported voice" })
+
+    with_stubbed_http(FakeResponse.new("400", body)) do |_captured, _response|
+      result = client.speak("Say hello")
+
+      assert_nil result["content"]
+      assert_nil result["response_id"]
+      assert_equal 400, result["status"]
+      assert_equal "Unsupported voice", result["error"]
+      assert_nil result["raw"]
+    end
+  end
+
   def test_http_errors_return_standard_envelope
     client = AiLite.new(api_key: "token-abc")
     body = JSON.generate("error" => { "message" => "Invalid API key" })
@@ -659,6 +919,30 @@ class AiLiteTest < Minitest::Test
     FakeResponse.new("200", body)
   end
 
+  def image_response(b64_json: Base64.strict_encode64("fake image"))
+    image = {}
+    image["b64_json"] = b64_json if b64_json
+
+    body = JSON.generate(
+      "created" => 1_713_833_628,
+      "background" => "opaque",
+      "data" => [image],
+      "output_format" => "png",
+      "quality" => "medium",
+      "size" => "1024x1024",
+      "usage" => {
+        "input_tokens" => 10,
+        "output_tokens" => 20,
+        "total_tokens" => 30
+      }
+    )
+    FakeResponse.new("200", body)
+  end
+
+  def speech_response(audio = "fake audio")
+    FakeResponse.new("200", audio)
+  end
+
   def with_env(values)
     originals = {}
 

metadata

@@ -1,18 +1,18 @@
 --- !ruby/object:Gem::Specification
 name: ai-lite
 version: !ruby/object:Gem::Version
-  version: 0.3.0
+  version: 0.5.0
 platform: ruby
 authors:
 - William Basmayor
-autorequire: 
+autorequire:
 bindir: bin
 cert_chain: []
-date: 2026-07-07 00:00:00.000000000 Z
+date: 2026-08-24 00:00:00.000000000 Z
 dependencies: []
 description: AI Lite is a dependency-light Ruby client for Rails apps and plain Ruby
   projects that need simple OpenAI Responses API calls.
-email: 
+email:
 executables: []
 extensions: []
 extra_rdoc_files: []
@@ -27,7 +27,7 @@ homepage: https://github.com/wbasmayor/ai-lite
 licenses:
 - MIT
 metadata: {}
-post_install_message: 
+post_install_message:
 rdoc_options: []
 require_paths:
 - lib
@@ -42,8 +42,8 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.0.3.1
-signing_key: 
+rubygems_version: 3.5.22
+signing_key:
 specification_version: 4
 summary: Minimal Ruby client for simple AI chat calls through the OpenAI Responses
   API.