ruby-openai

Ruby OpenAI

Use the OpenAI API with Ruby! 🤖❤️

Stream GPT-5 chats with the Responses API, initiate Realtime WebRTC conversations, and much more...

Sponsors

InferToGo - The inference addon for your PaaS application.

SerpApi - Search API - Enhance your LLM's knowledge with data from search engines like Google and Bing using our simple API.

🎮 Ruby AI Builders Discord | 🐦 X | 🧠 Anthropic Gem | 🚂 Midjourney Gem | ♥️ Thanks to all sponsors!

Contents

• Ruby OpenAI
  • Contents
  • Project Policies
  • Installation
    • Bundler
    • Gem install
  • How to use
    • Quickstart
    • With Config
      • Custom timeout or base URI
      • Extra Headers per Client
      • Logging
        • Errors
        • Faraday middleware
      • Azure
      • Deepseek
      • Ollama
      • Groq
      • Gemini
    • Counting Tokens
    • Models
    • Chat
      • Streaming Chat
      • Vision
      • JSON Mode
    • Responses API
      • Create a Response
      • Follow-up Messages
      • Tool Calls
      • Streaming
      • Retrieve a Response
      • Delete a Response
      • List Input Items
    • Functions
    • Completions
    • Embeddings
    • Batches
    • Files
      • For fine-tuning purposes
      • For assistant purposes
    • Finetunes
    • Vector Stores
    • Vector Store Files
    • Vector Store File Batches
    • Conversations
    • Assistants
    • Threads and Messages
    • Runs
      • Create and Run
      • Vision in a thread
      • Runs involving function tools
      • Exploring chunks used in File Search
    • Image Generation
      • DALL·E 2
      • DALL·E 3
    • Image Edit
    • Image Variations
    • Moderations
    • Whisper
      • Translate
      • Transcribe
      • Speech
    • Real-Time
    • Usage
    • Errors
  • Development
    • To check for deprecations
  • Release
  • Contributing
  • License
  • Code of Conduct

Project Policies

• Security policy
• Support and end-of-life policy
• Migration guides
• Changelog

Installation

Bundler

Add this line to your application's Gemfile:

gem "ruby-openai"

And then execute:

bundle install

Gem install

Or install with:

gem install ruby-openai

and require with:

require "openai"

How to use

• Get your API key from https://platform.openai.com/account/api-keys
• If you belong to multiple organizations, you can get your Organization ID from https://platform.openai.com/account/org-settings

Quickstart

For a quick test you can pass your token directly to a new client:

client = OpenAI::Client.new(
  access_token: "access_token_goes_here",
  log_errors: true # Highly recommended in development, so you can see what errors OpenAI is returning. Not recommended in production because it could leak private data to your logs.
)

With Config

For a more robust setup, you can configure the gem with your API keys, for example in an openai.rb initializer file. Never hardcode secrets into your codebase - instead use something like dotenv to pass the keys safely into your environments.

OpenAI.configure do |config|
  config.access_token = ENV.fetch("OPENAI_ACCESS_TOKEN")
  config.admin_token = ENV.fetch("OPENAI_ADMIN_TOKEN") # Optional, used for admin endpoints, created here: https://platform.openai.com/settings/organization/admin-keys
  config.organization_id = ENV.fetch("OPENAI_ORGANIZATION_ID") # Optional
  config.log_errors = true # Highly recommended in development, so you can see what errors OpenAI is returning. Not recommended in production because it could leak private data to your logs.
end

Then you can create a client like this:

client = OpenAI::Client.new

You can still override the config defaults when making new clients; any options not included will fall back to any global config set with OpenAI.configure. e.g. in this example the organization_id, request_timeout, etc. will fallback to any set globally using OpenAI.configure, with only the access_token and admin_token overridden:

client = OpenAI::Client.new(access_token: "access_token_goes_here", admin_token: "admin_token_goes_here")

Custom timeout or base URI

• The default timeout for any request using this library is 120 seconds. You can change that by passing a number of seconds to the request_timeout when initializing the client.
• You can also change the base URI used for all requests, eg. to use observability tools like Helicone or Velvet
• You can also add arbitrary other headers e.g. for openai-caching-proxy-worker, eg.:

client = OpenAI::Client.new(
  access_token: "access_token_goes_here",
  uri_base: "https://oai.hconeai.com/",
  request_timeout: 240,
  extra_headers: {
    "X-Proxy-TTL" => "43200", # For https://github.com/6/openai-caching-proxy-worker#specifying-a-cache-ttl
    "X-Proxy-Refresh": "true", # For https://github.com/6/openai-caching-proxy-worker#refreshing-the-cache
    "Helicone-Auth": "Bearer HELICONE_API_KEY", # For https://docs.helicone.ai/getting-started/integration-method/openai-proxy
    "helicone-stream-force-format" => "true", # Use this with Helicone otherwise streaming drops chunks # https://github.com/alexrudall/ruby-openai/issues/251
  }
)

or when configuring the gem:

OpenAI.configure do |config|
  config.access_token = ENV.fetch("OPENAI_ACCESS_TOKEN")
  config.admin_token = ENV.fetch("OPENAI_ADMIN_TOKEN") # Optional, used for admin endpoints, created here: https://platform.openai.com/settings/organization/admin-keys
  config.organization_id = ENV.fetch("OPENAI_ORGANIZATION_ID") # Optional
  config.log_errors = true # Optional
  config.uri_base = "https://oai.hconeai.com/" # Optional
  config.request_timeout = 240 # Optional
  config.extra_headers = {
    "X-Proxy-TTL" => "43200", # For https://github.com/6/openai-caching-proxy-worker#specifying-a-cache-ttl
    "X-Proxy-Refresh": "true", # For https://github.com/6/openai-caching-proxy-worker#refreshing-the-cache
    "Helicone-Auth": "Bearer HELICONE_API_KEY" # For https://docs.helicone.ai/getting-started/integration-method/openai-proxy
  } # Optional
end

Extra Headers per Client

You can dynamically pass headers per client object, which will be merged with any headers set globally with OpenAI.configure:

client = OpenAI::Client.new(access_token: "access_token_goes_here")
client.add_headers("X-Proxy-TTL" => "43200")

Logging

Errors

By default, ruby-openai does not log any Faraday::Errors encountered while executing a network request to avoid leaking data (e.g. 400s, 500s, SSL errors and more - see here for a complete list of subclasses of Faraday::Error and what can cause them).

If you would like to enable this functionality, you can set log_errors to true when configuring the client:

client = OpenAI::Client.new(log_errors: true)

Faraday middleware

You can pass Faraday middleware to the client in a block, eg:

• To enable verbose logging with Ruby's Logger:

client = OpenAI::Client.new do |f|
  f.response :logger, Logger.new($stdout), bodies: true
end

• To add a web debugging proxy like Charles:

  client = OpenAI::Client.new do |f|
    f.proxy = { uri: "http://localhost:8888" }
  end

Azure

To use the Azure OpenAI Service API, you can configure the gem like this:

OpenAI.configure do |config|
  config.access_token = ENV.fetch("AZURE_OPENAI_API_KEY")
  config.uri_base = ENV.fetch("AZURE_OPENAI_URI")
  config.api_type = :azure
  config.api_version = "2023-03-15-preview"
end

where AZURE_OPENAI_URI is e.g. https://custom-domain.openai.azure.com/openai/deployments/gpt-35-turbo

Deepseek

Deepseek is compatible with the OpenAI chat API. Get an access token from here, then:

client = OpenAI::Client.new(
  access_token: "deepseek_access_token_goes_here",
  uri_base: "https://api.deepseek.com/"
)

client.chat(
  parameters: {
    model: "deepseek-chat", # Required.
    messages: [{ role: "user", content: "Hello!"}], # Required.
    temperature: 0.7,
    stream: proc do |chunk, _event|
     print chunk.dig("choices", 0, "delta", "content")
    end
  }
)

Ollama

Ollama allows you to run open-source LLMs, such as Llama 3, locally. It offers chat compatibility with the OpenAI API.

You can download Ollama here. On macOS you can install and run Ollama like this:

brew install ollama
ollama serve
ollama pull llama3:latest # In new terminal tab.

Create a client using your Ollama server and the pulled model, and stream a conversation for free:

client = OpenAI::Client.new(
  uri_base: "http://localhost:11434"
)

client.chat(
  parameters: {
    model: "llama3", # Required.
    messages: [{ role: "user", content: "Hello!"}], # Required.
    temperature: 0.7,
    stream: proc do |chunk, _event|
      print chunk.dig("choices", 0, "delta", "content")
    end
  }
)

# => Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat?

Groq

Groq API Chat is broadly compatible with the OpenAI API, with a few minor differences. Get an access token from here, then:

client = OpenAI::Client.new(
  access_token: "groq_access_token_goes_here",
  uri_base: "https://api.groq.com/openai"
)

client.chat(
  parameters: {
    model: "llama3-8b-8192", # Required.
    messages: [{ role: "user", content: "Hello!"}], # Required.
    temperature: 0.7,
    stream: proc do |chunk, _event|
     print chunk.dig("choices", 0, "delta", "content")
    end
  }
)

Gemini

Gemini API Chat is also broadly compatible with the OpenAI API, and currently in beta. Get an access token from here, then:

client = OpenAI::Client.new(
  access_token: "gemini_access_token_goes_here",
  uri_base: "https://generativelanguage.googleapis.com/v1beta/openai/"
)

client.chat(
  parameters: {
    model: "gemini-1.5-flash", # Required.
    messages: [{ role: "user", content: "Hello!"}], # Required.
    temperature: 0.7,
    stream: proc do |chunk, _bytesize|
     print chunk.dig("choices", 0, "delta", "content")
    end
  }
)

# => Hello there! How can I help you today?

Counting Tokens

OpenAI parses prompt text into tokens, which are words or portions of words. (These tokens are unrelated to your API access_token.) Counting tokens can help you estimate your costs. It can also help you ensure your prompt text size is within the max-token limits of your model's context window, and choose an appropriate max_tokens completion parameter so your response will fit as well.

To estimate the token-count of your text:

OpenAI.rough_token_count("Your text")

If you need a more accurate count, try tiktoken_ruby.

Models

There are different models that can be used to generate text. For a full list and to retrieve information about a single model:

client.models.list
client.models.retrieve(id: "gpt-4o")

You can also delete any finetuned model you generated, if you're an account Owner on your OpenAI organization:

client.models.delete(id: "ft:gpt-4o-mini:acemeco:suffix:abc123")

Chat

GPT is a model that can be used to generate text in a conversational style. You can use it to generate a response to a sequence of messages:

response = client.chat(
  parameters: {
    model: "gpt-4o", # Required.
    messages: [{ role: "user", content: "Hello!"}], # Required.
    temperature: 0.7,
  }
)
puts response.dig("choices", 0, "message", "content")
# => "Hello! How may I assist you today?"

Streaming Chat

Quick guide to streaming Chat with Rails 7 and Hotwire

You can stream from the API in realtime, which can be much faster and used to create a more engaging user experience. Pass a Proc (or any object with a #call method) to the stream parameter to receive the stream of completion chunks as they are generated. Each time one or more chunks is received, the proc will be called once with each chunk, parsed as a Hash. If OpenAI returns an error, ruby-openai will raise a Faraday error.

client.chat(
  parameters: {
    model: "gpt-4o", # Required.
    messages: [{ role: "user", content: "Describe a character called Anna!"}], # Required.
    temperature: 0.7,
    stream: proc do |chunk, _event|
      print chunk.dig("choices", 0, "delta", "content")
    end
  }
)
# => "Anna is a young woman in her mid-twenties, with wavy chestnut hair that falls to her shoulders..."

Note: In order to get usage information, you can provide the stream_options parameter and OpenAI will provide a final chunk with the usage. Here is an example:

stream_proc = proc { |chunk, _bytesize| puts "--------------"; puts chunk.inspect; }
client.chat(
  parameters: {
    model: "gpt-4o",
    stream: stream_proc,
    stream_options: { include_usage: true },
    messages: [{ role: "user", content: "Hello!"}],
  }
)
# => --------------
# => {"id"=>"chatcmpl-7bbq05PiZqlHxjV1j7OHnKKDURKaf", "object"=>"chat.completion.chunk", "created"=>1718750612, "model"=>"gpt-4o-2024-05-13", "system_fingerprint"=>"fp_9cb5d38cf7", "choices"=>[{"index"=>0, "delta"=>{"role"=>"assistant", "content"=>""}, "logprobs"=>nil, "finish_reason"=>nil}], "usage"=>nil}
# => --------------
# => {"id"=>"chatcmpl-7bbq05PiZqlHxjV1j7OHnKKDURKaf", "object"=>"chat.completion.chunk", "created"=>1718750612, "model"=>"gpt-4o-2024-05-13", "system_fingerprint"=>"fp_9cb5d38cf7", "choices"=>[{"index"=>0, "delta"=>{"content"=>"Hello"}, "logprobs"=>nil, "finish_reason"=>nil}], "usage"=>nil}
# => --------------
# => ... more content chunks
# => --------------
# => {"id"=>"chatcmpl-7bbq05PiZqlHxjV1j7OHnKKDURKaf", "object"=>"chat.completion.chunk", "created"=>1718750612, "model"=>"gpt-4o-2024-05-13", "system_fingerprint"=>"fp_9cb5d38cf7", "choices"=>[{"index"=>0, "delta"=>{}, "logprobs"=>nil, "finish_reason"=>"stop"}], "usage"=>nil}
# => --------------
# => {"id"=>"chatcmpl-7bbq05PiZqlHxjV1j7OHnKKDURKaf", "object"=>"chat.completion.chunk", "created"=>1718750612, "model"=>"gpt-4o-2024-05-13", "system_fingerprint"=>"fp_9cb5d38cf7", "choices"=>[], "usage"=>{"prompt_tokens"=>9, "completion_tokens"=>9, "total_tokens"=>18}}

Vision

You can use the GPT-4 Vision model to generate a description of an image:

messages = [
  { "type": "text", "text": "What’s in this image?"},
  { "type": "image_url",
    "image_url": {
      "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg",
    },
  }
]
response = client.chat(
  parameters: {
    model: "gpt-4-vision-preview", # Required.
    messages: [{ role: "user", content: messages}], # Required.
  }
)
puts response.dig("choices", 0, "message", "content")
# => "The image depicts a serene natural landscape featuring a long wooden boardwalk extending straight ahead"

JSON Mode

You can set the response_format to ask for responses in JSON:

response = client.chat(
  parameters: {
    model: "gpt-4o",
    response_format: { type: "json_object" },
    messages: [{ role: "user", content: "Hello! Give me some JSON please."}],
    temperature: 0.7,
  })
  puts response.dig("choices", 0, "message", "content")
  # =>
  # {
  #   "name": "John",
  #   "age": 30,
  #   "city": "New York",
  #   "hobbies": ["reading", "traveling", "hiking"],
  #   "isStudent": false
  # }

You can stream it as well!

  response = client.chat(
    parameters: {
      model: "gpt-4o",
      messages: [{ role: "user", content: "Can I have some JSON please?"}],
      response_format: { type: "json_object" },
      stream: proc do |chunk, _event|
        print chunk.dig("choices", 0, "delta", "content")
      end
    }
  )
  # =>
  # {
  #   "message": "Sure, please let me know what specific JSON data you are looking for.",
  #   "JSON_data": {
  #     "example_1": {
  #       "key_1": "value_1",
  #       "key_2": "value_2",
  #       "key_3": "value_3"
  #     },
  #     "example_2": {
  #       "key_4": "value_4",
  #       "key_5": "value_5",
  #       "key_6": "value_6"
  #     }
  #   }
  # }

Responses API

OpenAI's most advanced interface for generating model responses. Supports text and image inputs, and text outputs. Create stateful interactions with the model, using the output of previous responses as input. Extend the model's capabilities with built-in tools for file search, web search, computer use, and more. Allow the model access to external systems and data using function calling.

Create a Response

response = client.responses.create(parameters: {
  model: "gpt-5",
  input: "Hello! I'm Szymon!",
  reasoning: {
    "effort": "minimal"
  }
})
puts response.dig("output", 0, "content", 0, "text")
# => Thinking about how to answer this...
puts response.dig("output", 1, "content", 0, "text")
# => Hi Szymon! Great to meet you. How can I help today?

Follow-up Messages

followup = client.responses.create(parameters: {
  model: "gpt-4o",
  input: "Remind me, what is my name?",
  previous_response_id: response["id"]
})
puts followup.dig("output", 0, "content", 0, "text")
# => Your name is Szymon! How can I help you today?

Tool Calls

response = client.responses.create(parameters: {
  model: "gpt-4o",
  input: "What's the weather in Paris?",
  tools: [
    {
      "type" => "function",
      "name" => "get_current_weather",
      "description" => "Get the current weather in a given location",
      "parameters" => {
        "type" => "object",
        "properties" => {
          "location" => {
            "type" => "string",
            "description" => "The geographic location to get the weather for"
          }
        },
        "required" => ["location"]
      }
    }
  ]
})
puts response.dig("output", 0, "name")
# => "get_current_weather"

Streaming

client.responses.create(
  parameters: {
    model: "gpt-4o", # Required.
    input: "Hello!", # Required.
    stream: proc do |chunk, _event|
      if chunk["type"] == "response.output_text.delta"
        print chunk["delta"]
        $stdout.flush  # Ensure output is displayed immediately
      end
    end
  }
)
# => "Hi there! How can I assist you today?..."

Retrieve a Response

retrieved_response = client.responses.retrieve(response_id: response["id"])
puts retrieved_response["object"]
# => "response"

Delete a Response

deletion = client.responses.delete(response_id: response["id"])
puts deletion["deleted"]
# => true

List Input Items

input_items = client.responses.input_items(response_id: response["id"])
puts input_items["object"] # => "list"

Functions

You can describe and pass in functions and the model will intelligently choose to output a JSON object containing arguments to call them - eg., to use your method get_current_weather to get the weather in a given location. Note that tool_choice is optional, but if you exclude it, the model will choose whether to use the function or not (see here).

def get_current_weather(location:, unit: "fahrenheit")
  # Here you could use a weather api to fetch the weather.
  "The weather in #{location} is nice 🌞 #{unit}"
end

messages = [
  {
    "role": "user",
    "content": "What is the weather like in San Francisco?",
  },
]

response =
  client.chat(
    parameters: {
      model: "gpt-4o",
      messages: messages,  # Defined above because we'll use it again
      tools: [
        {
          type: "function",
          function: {
            name: "get_current_weather",
            description: "Get the current weather in a given location",
            parameters: {  # Format: https://json-schema.org/understanding-json-schema
              type: :object,
              properties: {
                location: {
                  type: :string,
                  description: "The city and state, e.g. San Francisco, CA",
                },
                unit: {
                  type: "string",
                  enum: %w[celsius fahrenheit],
                },
              },
              required: ["location"],
            },
          },
        }
      ],
      # Optional, defaults to "auto"
      # Can also put "none" or specific functions, see docs
      tool_choice: "required"
    },
  )

message = response.dig("choices", 0, "message")

if message["role"] == "assistant" && message["tool_calls"]
  # For a subsequent message with the role "tool", OpenAI requires the preceding message to have a single tool_calls argument.
  messages << message

  message["tool_calls"].each do |tool_call|
    tool_call_id = tool_call.dig("id")
    function_name = tool_call.dig("function", "name")
    function_args = JSON.parse(
      tool_call.dig("function", "arguments"),
      { symbolize_names: true },
    )
    function_response =
      case function_name
      when "get_current_weather"
        get_current_weather(**function_args)  # => "The weather is nice 🌞"
      else
        # decide how to handle
      end

    messages << {
      tool_call_id: tool_call_id,
      role: "tool",
      name: function_name,
      content: function_response
    }  # Extend the conversation with the results of the functions
  end

  second_response = client.chat(
    parameters: {
      model: "gpt-4o",
      messages: messages
    }
  )

  puts second_response.dig("choices", 0, "message", "content")

  # At this point, the model has decided to call functions, you've called the functions
  # and provided the response back, and the model has considered this and responded.
end
# => "It looks like the weather is nice and sunny in San Francisco! If you're planning to go out, it should be a pleasant day."

Completions

Hit the OpenAI API for a completion using other GPT-3 models:

response = client.completions(
  parameters: {
    model: "gpt-4o",
    prompt: "Once upon a time",
    max_tokens: 5
  }
)
puts response["choices"].map { |c| c["text"] }
# => [", there lived a great"]

Embeddings

You can use the embeddings endpoint to get a vector of numbers representing an input. You can then compare these vectors for different inputs to efficiently check how similar the inputs are.

response = client.embeddings(
  parameters: {
    model: "text-embedding-ada-002",
    input: "The food was delicious and the waiter..."
  }
)

puts response.dig("data", 0, "embedding")
# => Vector representation of your embedding

Batches

The Batches endpoint allows you to create and manage large batches of API requests to run asynchronously. Currently, the supported endpoints for batches are /v1/chat/completions (Chat Completions API) and /v1/embeddings (Embeddings API).

To use the Batches endpoint, you need to first upload a JSONL file containing the batch requests using the Files endpoint. The file must be uploaded with the purpose set to batch. Each line in the JSONL file represents a single request and should have the following format:

{
  "custom_id": "request-1",
  "method": "POST",
  "url": "/v1/chat/completions",
  "body": {
    "model": "gpt-4o",
    "messages": [
      { "role": "system", "content": "You are a helpful assistant." },
      { "role": "user", "content": "What is 2+2?" }
    ]
  }
}

Once you have uploaded the JSONL file, you can create a new batch by providing the file ID, endpoint, and completion window:

response = client.batches.create(
  parameters: {
    input_file_id: "file-abc123",
    endpoint: "/v1/chat/completions",
    completion_window: "24h"
  }
)
batch_id = response["id"]

You can retrieve information about a specific batch using its ID:

batch = client.batches.retrieve(id: batch_id)

To cancel a batch that is in progress:

client.batches.cancel(id: batch_id)

You can also list all the batches:

client.batches.list

Once the batch["completed_at"] is present, you can fetch the output or error files:

batch = client.batches.retrieve(id: batch_id)
output_file_id = batch["output_file_id"]
output_response = client.files.content(id: output_file_id)
error_file_id = batch["error_file_id"]
error_response = client.files.content(id: error_file_id)

These files are in JSONL format, with each line representing the output or error for a single request. The lines can be in any order:

{
  "id": "response-1",
  "custom_id": "request-1",
  "response": {
    "id": "chatcmpl-abc123",
    "object": "chat.completion",
    "created": 1677858242,
    "model": "gpt-4o",
    "choices": [
      {
        "index": 0,
        "message": {
          "role": "assistant",
          "content": "2+2 equals 4."
        }
      }
    ]
  }
}

If a request fails with a non-HTTP error, the error object will contain more information about the cause of the failure.

Files

For fine-tuning purposes

Put your data in a .jsonl file like this:

{"prompt":"Overjoyed with my new phone! ->", "completion":" positive"}
{"prompt":"@lakers disappoint for a third straight night ->", "completion":" negative"}

and pass the path (or a StringIO object) to client.files.upload to upload it to OpenAI, and then interact with it:

client.files.upload(parameters: { file: "path/to/sentiment.jsonl", purpose: "fine-tune" })
client.files.list
client.files.retrieve(id: "file-123")
client.files.content(id: "file-123")
client.files.delete(id: "file-123")

For assistant purposes

You can send a file path:

client.files.upload(parameters: { file: "path/to/file.pdf", purpose: "assistants" })

or a File object

my_file = File.open("path/to/file.pdf", "rb")
client.files.upload(parameters: { file: my_file, purpose: "assistants" })

See supported file types on API documentation.

Finetunes

Upload your fine-tuning data in a .jsonl file as above and get its ID:

response = client.files.upload(parameters: { file: "path/to/sarcasm.jsonl", purpose: "fine-tune" })
file_id = JSON.parse(response.body)["id"]

You can then use this file ID to create a fine tuning job:

response = client.finetunes.create(
  parameters: {
  training_file: file_id,
  model: "gpt-4o"
})
fine_tune_id = response["id"]

That will give you the fine-tune ID. If you made a mistake you can cancel the fine-tune model before it is processed:

client.finetunes.cancel(id: fine_tune_id)

You may need to wait a short time for processing to complete. Once processed, you can use list or retrieve to get the name of the fine-tuned model:

client.finetunes.list
response = client.finetunes.retrieve(id: fine_tune_id)
fine_tuned_model = response["fine_tuned_model"]

This fine-tuned model name can then be used in chat completions:

response = client.chat(
  parameters: {
    model: fine_tuned_model,
    messages: [{ role: "user", content: "I love Mondays!" }]
  }
)
response.dig("choices", 0, "message", "content")

You can also capture the events for a job:

client.finetunes.list_events(id: fine_tune_id)

You can also delete any finetuned model you generated, if you're an account Owner on your OpenAI organization:

client.models.delete(id: fine_tune_id)

Vector Stores

Vector Store objects give the File Search tool the ability to search your files.

You can create a new vector store:

response = client.vector_stores.create(
  parameters: {
    name: "my vector store",
    file_ids: ["file-abc123", "file-def456"]
  }
)

vector_store_id = response["id"]

Given a vector_store_id you can retrieve the current field values:

client.vector_stores.retrieve(id: vector_store_id)

You can get a list of all vector stores currently available under the organization:

client.vector_stores.list

You can modify an existing vector store, except for the file_ids:

response = client.vector_stores.modify(
  id: vector_store_id,
  parameters: {
    name: "Modified Test Vector Store",
  }
)

You can search a vector store for relevant chunks based on a query:

response = client.vector_stores.search(
  id: vector_store_id,
  parameters: {
    query: "What is the return policy?",
    max_num_results: 20,
    ranking_options: {
      # Add any ranking options here in line with the API documentation
    },
    rewrite_query: true,
    filters: {
      type: "eq",
      property: "region",
      value: "us"
    }
  }
)

You can delete vector stores:

client.vector_stores.delete(id: vector_store_id)

Vector Store Files

Vector store files represent files inside a vector store.

You can create a new vector store file by attaching a File to a vector store.

response = client.vector_store_files.create(
  vector_store_id: "vector-store-abc123",
  parameters: {
    file_id: "file-abc123"
  }
)

vector_store_file_id = response["id"]

Given a vector_store_file_id you can retrieve the current field values:

client.vector_store_files.retrieve(
  vector_store_id: "vector-store-abc123",
  id: vector_store_file_id
)

You can get a list of all vector store files currently available under the vector store:

client.vector_store_files.list(vector_store_id: "vector-store-abc123")

You can delete a vector store file:

client.vector_store_files.delete(
  vector_store_id: "vector-store-abc123",
  id: vector_store_file_id
)

Note: This will remove the file from the vector store but the file itself will not be deleted. To delete the file, use the delete file endpoint.

Vector Store File Batches

Vector store file batches represent operations to add multiple files to a vector store.

You can create a new vector store file batch by attaching multiple Files to a vector store.

response = client.vector_store_file_batches.create(
  vector_store_id: "vector-store-abc123",
  parameters: {
    file_ids: ["file-abc123", "file-def456"]
  }
)

file_batch_id = response["id"]

Given a file_batch_id you can retrieve the current field values:

client.vector_store_file_batches.retrieve(
  vector_store_id: "vector-store-abc123",
  id: file_batch_id
)

You can get a list of all vector store files in a batch currently available under the vector store:

client.vector_store_file_batches.list(
  vector_store_id: "vector-store-abc123",
  id: file_batch_id
)

You can cancel a vector store file batch (This attempts to cancel the processing of files in this batch as soon as possible):

client.vector_store_file_batches.cancel(
  vector_store_id: "vector-store-abc123",
  id: file_batch_id
)

Conversations

The Conversations API enables you to create and manage persistent conversations with your models. This is useful for maintaining conversation state across multiple interactions.

Supported Endpoints:

• POST /v1/conversations - Create a conversation
• GET /v1/conversations/{id} - Retrieve a conversation
• PATCH /v1/conversations/{id} - Modify a conversation
• DELETE /v1/conversations/{id} - Delete a conversation
• POST /v1/conversations/{id}/items - Create items in a conversation
• GET /v1/conversations/{id}/items - List items in a conversation
• GET /v1/conversations/{id}/items/{item_id} - Get a specific item
• DELETE /v1/conversations/{id}/items/{item_id} - Delete an item

Creating a Conversation

To create a new conversation:

response = client.conversations.create(
  parameters: {
    metadata: { purpose: "customer_support" }
  }
)
conversation_id = response["id"]

Retrieving a Conversation

To retrieve a specific conversation:

conversation = client.conversations.retrieve(id: conversation_id)

Modifying a Conversation

To update a conversation's metadata:

response = client.conversations.modify(
  id: conversation_id,
  parameters: {
    metadata: { status: "resolved" }
  }
)

Deleting a Conversation

To delete a conversation:

response = client.conversations.delete(id: conversation_id)

Managing Items in Conversations

You can add, retrieve, and manage items within a conversation.

Creating Items

# Create multiple items at once
response = client.conversations.create_items(
  conversation_id: conversation_id,
  parameters: {
    items: [
      {
        type: "message",
        role: "user",
        content: [
          { type: "input_text", text: "Hello!" }
        ]
      },
      {
        type: "message",
        role: "assistant",
        content: [
          { type: "input_text", text: "How are you?" }
        ]
      }
    ]
  }
)

Listing Items

# List all items in a conversation
response = client.conversations.list_items(conversation_id: conversation_id)
items = response["data"]

# With parameters
response = client.conversations.list_items(
  conversation_id: conversation_id,
  parameters: {
    limit: 10,
    order: "asc"
  }
)

Retrieving a Specific Item

item = client.conversations.get_item(
  conversation_id: conversation_id,
  item_id: item_id
)

Deleting an Item

response = client.conversations.delete_item(
  conversation_id: conversation_id,
  item_id: item_id
)

Assistants

Assistants are stateful actors that can have many conversations and use tools to perform tasks (see Assistant Overview).

To create a new assistant:

response = client.assistants.create(
  parameters: {
    model: "gpt-4o",
    name: "OpenAI-Ruby test assistant",
    description: nil,
    instructions: "You are a Ruby dev bot. When asked a question, write and run Ruby code to answer the question",
    tools: [
      { type: "code_interpreter" },
      { type: "file_search" }
    ],
    tool_resources: {
      code_interpreter: {
        file_ids: [] # See Files section above for how to upload files
      },
      file_search: {
        vector_store_ids: [] # See Vector Stores section above for how to add vector stores
      }
    },
    "metadata": { my_internal_version_id: "1.0.0" }
  }
)
assistant_id = response["id"]

Given an assistant_id you can retrieve the current field values:

client.assistants.retrieve(id: assistant_id)

You can get a list of all assistants currently available under the organization:

client.assistants.list

You can modify an existing assistant using the assistant's id (see API documentation):

response = client.assistants.modify(
  id: assistant_id,
  parameters: {
    name: "Modified Test Assistant for OpenAI-Ruby",
    metadata: { my_internal_version_id: '1.0.1' }
  }
)

You can delete assistants:

client.assistants.delete(id: assistant_id)

Threads and Messages

Once you have created an assistant as described above, you need to prepare a Thread of Messages for the assistant to work on (see introduction on Assistants). For example, as an initial setup you could do:

# Create thread
response = client.threads.create # Note: Once you create a thread, there is no way to list it
                                 # or recover it currently (as of 2023-12-10). So hold onto the `id`
thread_id = response["id"]

# Add initial message from user (see https://platform.openai.com/docs/api-reference/messages/createMessage)
message_id = client.messages.create(
  thread_id: thread_id,
  parameters: {
    role: "user", # Required for manually created messages
    content: "Can you help me write an API library to interact with the OpenAI API please?"
  }
)["id"]

# Retrieve individual message
message = client.messages.retrieve(thread_id: thread_id, id: message_id)

# Review all messages on the thread
messages = client.messages.list(thread_id: thread_id)

To clean up after a thread is no longer needed:

# To delete the thread (and all associated messages):
client.threads.delete(id: thread_id)

client.messages.retrieve(thread_id: thread_id, id: message_id) # -> Fails after thread is deleted

Runs

To submit a thread to be evaluated with the model of an assistant, create a Run as follows:

# Create run (will use instruction/model/tools from Assistant's definition)
response = client.runs.create(
  thread_id: thread_id,
  parameters: {
    assistant_id: assistant_id,
    max_prompt_tokens: 256,
    max_completion_tokens: 16
  }
)
run_id = response['id']

You can stream the message chunks as they come through:

client.runs.create(
  thread_id: thread_id,
  parameters: {
    assistant_id: assistant_id,
    max_prompt_tokens: 256,
    max_completion_tokens: 16,
    stream: proc do |chunk, _event|
      if chunk["object"] == "thread.message.delta"
        print chunk.dig("delta", "content", 0, "text", "value")
      end
    end
  }
)

To get the status of a Run:

response = client.runs.retrieve(id: run_id, thread_id: thread_id)
status = response['status']

The status response can include the following strings queued, in_progress, requires_action, cancelling, cancelled, failed, completed, or expired which you can handle as follows:

while true do
  response = client.runs.retrieve(id: run_id, thread_id: thread_id)
  status = response['status']

  case status
  when 'queued', 'in_progress', 'cancelling'
    puts 'Sleeping'
    sleep 1 # Wait one second and poll again
  when 'completed'
    break # Exit loop and report result to user
  when 'requires_action'
    # Handle tool calls (see below)
  when 'cancelled', 'failed', 'expired'
    puts response['last_error'].inspect
    break # or `exit`
  else
    puts "Unknown status response: #{status}"
  end
end

If the status response indicates that the run is completed, the associated thread will have one or more new messages attached:

# Either retrieve all messages in bulk again, or...
messages = client.messages.list(thread_id: thread_id, parameters: { order: 'asc' })

# Alternatively retrieve the `run steps` for the run which link to the messages:
run_steps = client.run_steps.list(thread_id: thread_id, run_id: run_id, parameters: { order: 'asc' })
new_message_ids = run_steps['data'].filter_map do |step|
  if step['type'] == 'message_creation'
    step.dig('step_details', "message_creation", "message_id")
  end # Ignore tool calls, because they don't create new messages.
end

# Retrieve the individual messages
new_messages = new_message_ids.map do |msg_id|
  client.messages.retrieve(id: msg_id, thread_id: thread_id)
end

# Find the actual response text in the content array of the messages
new_messages.each do |msg|
  msg['content'].each do |content_item|
    case content_item['type']
    when 'text'
      puts content_item.dig('text', 'value')
      # Also handle annotations
    when 'image_file'
      # Use File endpoint to retrieve file contents via id
      id = content_item.dig('image_file', 'file_id')
    end
  end
end

You can also update the metadata on messages, including messages that come from the assistant.

metadata = {
  user_id: "abc123"
}
message = client.messages.modify(
  id: message_id,
  thread_id: thread_id,
  parameters: { metadata: metadata },
)

At any time you can list all runs which have been performed on a particular thread or are currently running:

client.runs.list(thread_id: thread_id, parameters: { order: "asc", limit: 3 })

Create and Run

You can also create a thread and run in one call like this:

response = client.runs.create_thread_and_run(parameters: { assistant_id: assistant_id })
run_id = response['id']
thread_id = response['thread_id']

Vision in a thread

You can include images in a thread and they will be described & read by the LLM. In this example I'm using this file:

require "openai"

# Make a client
client = OpenAI::Client.new(
  access_token: "access_token_goes_here",
  log_errors: true # Don't log errors in production.
)

# Upload image as a file
file_id = client.files.upload(
  parameters: {
    file: "path/to/example.png",
    purpose: "assistants",
  }
)["id"]

# Create assistant (You could also use an existing one here)
assistant_id = client.assistants.create(
  parameters: {
    model: "gpt-4o",
    name: "Image reader",
    instructions: "You are an image describer. You describe the contents of images.",
  }
)["id"]

# Create thread
thread_id = client.threads.create["id"]

# Add image in message
client.messages.create(
  thread_id: thread_id,
  parameters: {
    role: "user", # Required for manually created messages
    content: [
      {
        "type": "text",
        "text": "What's in this image?"
      },
      {
        "type": "image_file",
        "image_file": { "file_id": file_id }
      }
    ]
  }
)

# Run thread
run_id = client.runs.create(
  thread_id: thread_id,
  parameters: { assistant_id: assistant_id }
)["id"]

# Wait until run in complete
status = nil
until status == "completed" do
  sleep(0.1)
  status = client.runs.retrieve(id: run_id, thread_id: thread_id)['status']
end

# Get the response
messages = client.messages.list(thread_id: thread_id, parameters: { order: 'asc' })
messages.dig("data", -1, "content", 0, "text", "value")
=> "The image contains a placeholder graphic with a tilted, stylized representation of a postage stamp in the top part, which includes an abstract landscape with hills and a sun. Below the stamp, in the middle of the image, there is italicized text in a light golden color that reads, \"This is just an example.\" The background is a light pastel shade, and a yellow border frames the entire image."

Runs involving function tools

In case you are allowing the assistant to access function tools (they are defined in the same way as functions during chat completion), you might get a status code of requires_action when the assistant wants you to evaluate one or more function tools:

def get_current_weather(location:, unit: "celsius")
  # Your function code goes here
  if location =~ /San Francisco/i
    return unit == "celsius" ? "The weather is nice 🌞 at 27°C" : "The weather is nice 🌞 at 80°F"
  else
    return unit == "celsius" ? "The weather is icy 🥶 at -5°C" : "The weather is icy 🥶 at 23°F"
  end
end

if status == 'requires_action'
  tools_to_call = response.dig('required_action', 'submit_tool_outputs', 'tool_calls')

  my_tool_outputs = tools_to_call.map { |tool|
    # Call the functions based on the tool's name
    function_name = tool.dig('function', 'name')
    arguments = JSON.parse(
      tool.dig("function", "arguments"),
      { symbolize_names: true },
    )

    tool_output = case function_name
    when "get_current_weather"
      get_current_weather(**arguments)
    end

    {
      tool_call_id: tool['id'],
      output: tool_output,
    }
  }

  client.runs.submit_tool_outputs(
    thread_id: thread_id,
    run_id: run_id,
    parameters: { tool_outputs: my_tool_outputs }
  )
end

Note that you have 10 minutes to submit your tool output before the run expires.

Exploring chunks used in File Search

Take a deep breath. You might need a drink for this one.

It's possible for OpenAI to share what chunks it used in its internal RAG Pipeline to create its filesearch results.

An example spec can be found here that does this, just so you know it's possible.

Here's how to get the chunks used in a file search. In this example I'm using this file:

require "openai"

# Make a client
client = OpenAI::Client.new(
  access_token: "access_token_goes_here",
  log_errors: true # Don't log errors in production.
)

# Upload your file(s)
file_id = client.files.upload(
  parameters: {
    file: "path/to/somatosensory.pdf",
    purpose: "assistants"
  }
)["id"]

# Create a vector store to store the vectorised file(s)
vector_store_id = client.vector_stores.create(parameters: {})["id"]

# Vectorise the file(s)
vector_store_file_id = client.vector_store_files.create(
  vector_store_id: vector_store_id,
  parameters: { file_id: file_id }
)["id"]

# Check that the file is vectorised (wait for status to be "completed")
client.vector_store_files.retrieve(vector_store_id: vector_store_id, id: vector_store_file_id)["status"]

# Create an assistant, referencing the vector store
assistant_id = client.assistants.create(
  parameters: {
    model: "gpt-4o",
    name: "Answer finder",
    instructions: "You are a file search tool. Find the answer in the given files, please.",
    tools: [
      { type: "file_search" }
    ],
    tool_resources: {
      file_search: {
        vector_store_ids: [vector_store_id]
      }
    }
  }
)["id"]

# Create a thread with your question
thread_id = client.threads.create(parameters: {
  messages: [
    { role: "user",
      content: "Find the description of a nociceptor." }
  ]
})["id"]

# Run the thread to generate the response. Include the "GIVE ME THE CHUNKS" incantation.
run_id = client.runs.create(
  thread_id: thread_id,
  parameters: {
    assistant_id: assistant_id
  },
  query_parameters: { include: ["step_details.tool_calls[*].file_search.results[*].content"] } # incantation
)["id"]

# Get the steps that happened in the run
steps = client.run_steps.list(
  thread_id: thread_id,
  run_id: run_id,
  parameters: { order: "asc" }
)

# Retrieve all the steps. Include the "GIVE ME THE CHUNKS" incantation again.
steps = steps["data"].map do |step|
  client.run_steps.retrieve(
    thread_id: thread_id,
    run_id: run_id,
    id: step["id"],
    parameters: { include: ["step_details.tool_calls[*].file_search.results[*].content"] } # incantation
  )
end

# Now we've got the chunk info, buried deep. Loop through the steps and find chunks if included:
chunks = steps.flat_map do |step|
  included_results = step.dig("step_details", "tool_calls", 0, "file_search", "results")

  next if included_results.nil? || included_results.empty?

  included_results.flat_map do |result|
    result["content"].map do |content|
      content["text"]
    end
  end
end.compact

# The first chunk will be the closest match to the prompt. Finally, if you want to view the completed message(s):
client.messages.list(thread_id: thread_id)

Image Generation

Generate images using DALL·E 2 or DALL·E 3!

DALL·E 2

For DALL·E 2 the size of any generated images must be one of 256x256, 512x512 or 1024x1024 - if not specified the image will default to 1024x1024.

response = client.images.generate(
  parameters: {
    prompt: "A baby sea otter cooking pasta wearing a hat of some sort",
    size: "256x256",
  }
)
puts response.dig("data", 0, "url")
# => "https://oaidalleapiprodscus.blob.core.windows.net/private/org-Rf437IxKhh..."

DALL·E 3

For DALL·E 3 the size of any generated images must be one of 1024x1024, 1024x1792 or 1792x1024. Additionally the quality of the image can be specified to either standard or hd.

response = client.images.generate(
  parameters: {
    prompt: "A springer spaniel cooking pasta wearing a hat of some sort",
    model: "dall-e-3",
    size: "1024x1792",
    quality: "standard",
  }
)
puts response.dig("data", 0, "url")
# => "https://oaidalleapiprodscus.blob.core.windows.net/private/org-Rf437IxKhh..."

Image Edit

Fill in the transparent part of an image, or upload a mask with transparent sections to indicate the parts of an image that can be changed according to your prompt...

response = client.images.edit(
  parameters: {
    prompt: "A solid red Ruby on a blue background",
    image: "image.png",
    mask: "mask.png",
  }
)
puts response.dig("data", 0, "url")
# => "https://oaidalleapiprodscus.blob.core.windows.net/private/org-Rf437IxKhh..."

You can also upload arrays of images, eg.

  client = OpenAI::Client.new
  response = client.images.edit(
    parameters: {
      model: "gpt-image-1",
      image: [File.open(base_image_path, "rb"), "image.png"],
      prompt: "Take the first image as base and apply the second image as a watermark on the bottom right corner",
      size: "1024x1024"
      # Removed response_format parameter as it's not supported with gpt-image-1
    }
  )

Image Variations

Create n variations of an image.

response = client.images.variations(parameters: { image: "image.png", n: 2 })
puts response.dig("data", 0, "url")
# => "https://oaidalleapiprodscus.blob.core.windows.net/private/org-Rf437IxKhh..."

Moderations

Pass a string to check if it violates OpenAI's Content Policy:

response = client.moderations(parameters: { input: "I'm worried about that." })
puts response.dig("results", 0, "category_scores", "hate")
# => 5.505014632944949e-05

Whisper

Whisper is a speech to text model that can be used to generate text based on audio files:

Translate

The translations API takes as input the audio file in any of the supported languages and transcribes the audio into English.

response = client.audio.translate(
  parameters: {
    model: "whisper-1",
    file: File.open("path_to_file", "rb"),
  }
)
puts response["text"]
# => "Translation of the text"

Transcribe

The transcriptions API takes as input the audio file you want to transcribe and returns the text in the desired output file format.

You can pass the language of the audio file to improve transcription quality. Supported languages are listed here. You need to provide the language as an ISO-639-1 code, eg. "en" for English or "ne" for Nepali. You can look up the codes here.

response = client.audio.transcribe(
  parameters: {
    model: "whisper-1",
    file: File.open("path_to_file", "rb"),
    language: "en", # Optional
  }
)
puts response["text"]
# => "Transcription of the text"

If you are using Ruby on Rails with Active Storage, you would need to send an audio or video file like this (User has_one_attached):

user.media.blob.open do |file|
  response = client.audio.transcribe(
    parameters: {
        model: "whisper-1",
        file: File.open(file, "rb"),
        language: "en" # Optional
    })
  puts response["text"]
  # => "Transcription of the text"
end

Speech

The speech API takes as input the text and a voice and returns the content of an audio file you can listen to.

response = client.audio.speech(
  parameters: {
    model: "tts-1",
    input: "This is a speech test!",
    voice: "alloy",
    response_format: "mp3", # Optional
    speed: 1.0, # Optional
  }
)
File.binwrite('demo.mp3', response)
# => mp3 file that plays: "This is a speech test!"

Realtime

The Realtime API allows you to create a live speech-to-speech session with an OpenAI model. It responds with a session object, plus a client_secret key which contains a usable ephemeral API token that can be used to authenticate browser clients for a WebRTC connection.

response = client.realtime.create(parameters: { model: "gpt-4o-realtime-preview-2024-12-17" })
puts "ephemeral key: #{response.dig('client_secret', 'value')}"
# => "ephemeral key: ek_abc123"

Then in the client-side Javascript application, make a POST request to the Real-Time API with the ephemeral key and the SDP offer.

const OPENAI_REALTIME_URL = 'https://api.openai.com/v1/realtime/sessions'
const MODEL = 'gpt-4o-realtime-preview-2024-12-17'

const response = await fetch(`${OPENAI_REALTIME_URL}?model=${MODEL}`, {
  method: 'POST',
  headers: {
    'Content-Type': 'application/sdp',
    'Authorization': `Bearer ${ephemeralKey}`,
    'OpenAI-Beta': 'realtime=v1'
  },
  body: offer.sdp
})

Usage

The Usage API provides information about the cost of various OpenAI services within your organization. To use Admin APIs like Usage, you need to set an OPENAI_ADMIN_TOKEN, which can be generated here.

OpenAI.configure do |config|
  config.admin_token = ENV.fetch("OPENAI_ADMIN_TOKEN")
end

# or

client = OpenAI::Client.new(admin_token: "123abc")

You can retrieve usage data for different endpoints and time periods:

one_day_ago = Time.now.to_i - 86_400

# Retrieve costs data
response = client.usage.costs(parameters: { start_time: one_day_ago })
response["data"].each do |bucket|
  bucket["results"].each do |result|
    puts "#{Time.at(bucket["start_time"]).to_date}: $#{result.dig("amount", "value").round(2)}"
  end
end
=> 2025-02-09: $0.0
=> 2025-02-10: $0.42

# Retrieve completions usage data
response = client.usage.completions(parameters: { start_time: one_day_ago })
puts response["data"]

# Retrieve embeddings usage data
response = client.usage.embeddings(parameters: { start_time: one_day_ago })
puts response["data"]

# Retrieve moderations usage data
response = client.usage.moderations(parameters: { start_time: one_day_ago })
puts response["data"]

# Retrieve image generation usage data
response = client.usage.images(parameters: { start_time: one_day_ago })
puts response["data"]

# Retrieve audio speech usage data
response = client.usage.audio_speeches(parameters: { start_time: one_day_ago })
puts response["data"]

# Retrieve audio transcription usage data
response = client.usage.audio_transcriptions(parameters: { start_time: one_day_ago })
puts response["data"]

# Retrieve vector stores usage data
response = client.usage.vector_stores(parameters: { start_time: one_day_ago })
puts response["data"]

Errors

HTTP errors can be caught like this:

begin
  OpenAI::Client.new.models.retrieve(id: "gpt-4o")
rescue Faraday::Error => e
  raise "Got a Faraday error: #{e}"
end

Development

After checking out the repo, run bin/setup to install dependencies. You can run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install.

To run all tests, execute the command bundle exec rake, which will also run the linter (Rubocop). This repository uses VCR to log API requests.

[!WARNING] If you have an OPENAI_ACCESS_TOKEN and OPENAI_ADMIN_TOKEN in your ENV, running the specs will hit the actual API, which will be slow and cost you money - 2 cents or more! Remove them from your environment with unset or similar if you just want to run the specs against the stored VCR responses.