Skip to main content

OpenAI

To use the OpenAI API, set the OPENAI_API_KEY environment variable, specify via apiKey field in the configuration file or pass the API key as an argument to the constructor.

Example:

export OPENAI_API_KEY=your_api_key_here

The OpenAI provider supports the following model formats:

  • openai:chat - defaults to gpt-4o-mini
  • openai:completion - defaults to text-davinci-003
  • openai:<model name> - uses a specific model name (mapped automatically to chat or completion endpoint)
  • openai:chat:<model name> - uses any model name against the /v1/chat/completions endpoint
  • openai:chat:ft:gpt-3.5-turbo-0613:company-name:ID - example of a fine-tuned chat completion model
  • openai:completion:<model name> - uses any model name against the /v1/completions endpoint
  • openai:embeddings:<model name> - uses any model name against the /v1/embeddings endpoint
  • openai:assistant:<assistant id> - use an assistant

The openai:<endpoint>:<model name> construction is useful if OpenAI releases a new model, or if you have a custom model. For example, if OpenAI releases gpt-5 chat completion, you could begin using it immediately with openai:chat:gpt-5.

The OpenAI provider supports a handful of configuration options, such as temperature, functions, and tools, which can be used to customize the behavior of the model like so:

providers:
  - id: openai:gpt-4o-mini
    config:
      temperature: 0
      max_tokens: 1024

Formatting chat messages​

For information on setting up chat conversation, see chat threads.

Configuring parameters​

The providers list takes a config key that allows you to set parameters like temperature, max_tokens, and others. For example:

providers:
  - id: openai:gpt-4o-mini
    config:
      temperature: 0
      max_tokens: 128
      apiKey: sk-abc123

Supported parameters include:

ParameterDescription
temperatureControls the randomness of the AI's output. Higher values (close to 1) make the output more random, while lower values (close to 0) make it more deterministic.
max_tokensControls the maximum length of the output in tokens.
top_pControls the nucleus sampling, a method that helps control the randomness of the AI's output.
frequency_penaltyApplies a penalty to frequent tokens, making them less likely to appear in the output.
presence_penaltyApplies a penalty to new tokens (tokens that haven't appeared in the input), making them less likely to appear in the output.
best_ofControls the number of alternative outputs to generate and select from.
functionsAllows you to define custom functions. Each function should be an object with a name, optional description, and parameters.
function_callControls whether the AI should call functions. Can be either 'none', 'auto', or an object with a name that specifies the function to call.
toolsAllows you to define custom tools. See OpenAI Tools documentation
tool_choiceControls whether the AI should use a tool. See OpenAI Tools documentation
stopDefines a list of tokens that signal the end of the output.
stopDefines a list of tokens that signal the end of the output.
response_formatResponse format restrictions.
seedSeed used for deterministic output. Defaults to 0
apiKeyYour OpenAI API key, equivalent to OPENAI_API_KEY environment variable
apiKeyEnvarAn environment variable that contains the API key
apiHostThe hostname of the OpenAI API, please also read OPENAI_API_HOST below.
apiBaseUrlThe base URL of the OpenAI API, please also read OPENAI_BASE_URL below.
organizationYour OpenAI organization key.
headersAdditional headers to include in the request.

Here are the type declarations of config parameters:

interface OpenAiConfig {
  // Completion parameters
  temperature?: number;
  max_tokens?: number;
  top_p?: number;
  frequency_penalty?: number;
  presence_penalty?: number;
  best_of?: number;
  functions?: {
    name: string;
    description?: string;
    parameters: any;
  }[];
  function_call?: 'none' | 'auto' | { name: string };
  tools?: {
    type: string;
    function: {
      name: string;
      description?: string;
      parameters: any;
    };
  }[];
  tool_choice?:
    | 'none'
    | 'auto'
    | { type: 'function'; function?: { name: string } }
    | { type: 'file_search' };
  stop?: string[];
  response_format?: { type: string };
  seed?: number;

  // General OpenAI parameters
  apiKey?: string;
  apiKeyEnvar?: string;
  apiHost?: string;
  apiBaseUrl?: string;
  organization?: string;
  headers?: { [key: string]: string };
}

Images​

Sending images in prompts​

You can include images in the prompt by using content blocks. For example, here's an example config:

prompts:
  - prompt.json

providers:
  - openai:gpt-4o

tests:
  - vars:
      question: 'What do you see?'
      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'
  # ...

And an example prompt.json:

[
  {
    "role": "user",
    "content": [
      {
        "type": "text",
        "text": "{{question}}"
      },
      {
        "type": "image_url",
        "image_url": {
          "url": "{{url}}"
        }
      }
    ]
  }
]

See the OpenAI vision example.

Generating images​

OpenAI supports Dall-E generations via openai:image:dalle-3. See the OpenAI Dall-E example.

prompts:
  - 'In the style of Van Gogh: {{subject}}'
  - 'In the style of Dali: {{subject}}'

providers:
  - openai:image:dall-e-3

tests:
  - vars:
      subject: bananas
  - vars:
      subject: new york city

To display images in the web viewer, wrap vars or outputs in markdown image tags like so:

![](/path/to/myimage.png)

Then, enable 'Render markdown' under Table Settings.

Using tools and functions​

OpenAI tools and functions are supported. See OpenAI tools example and OpenAI functions example.

Using tools​

To set tools on an OpenAI provider, use the provider's config key. Add your function definitions under this key.

prompts: [prompt.txt]
providers:
  - id: openai:chat:gpt-4o-mini
    config:
      tools: [
        {
        "type": "function",
          "function": {
            "name": "get_current_weather",
            "description": "Get the current weather in a given location",
            "parameters": {
              "type": "object",
                "properties": {
                  "location": {
                    "type": "string",
                      "description": "The city and state, e.g. San Francisco, CA"
                    },
                    "unit": {
                      "type": "string",
                      "enum": ["celsius", "fahrenheit"]
                    }
                  },
              "required": ["location"]
            }
          }
        }
      ]
      tool_choice: 'auto'

tests:
   - vars:
        city: Boston
     assert:
        - type: is-json
        - type: is-valid-openai-tools-call
        - type: javascript
          value: output[0].function.name === 'get_current_weather'
        - type: javascript
          value: JSON.parse(output[0].function.arguments).location === 'Boston, MA'

   - vars:
        city: New York
# ...

Sometimes OpenAI function calls don't match tools schemas. Use is-valid-openai-tools-call or is-valid-openai-tools-call assertions to enforce an exact schema match between tools and the function definition.

To further test tools definitions, you can use the javascript assertion and/or transform directives. For example:

tests:
  - vars:
      city: Boston
    assert:
      - type: is-json
      - type: is-valid-openai-tools-call
      - type: javascript
        value: output[0].function.name === 'get_current_weather'
      - type: javascript
        value: JSON.parse(output[0].function.arguments).location === 'Boston, MA'

  - vars:
      city: New York
      # transform returns only the 'name' property
    transform: output[0].function.name
    assert:
      - type: is-json
      - type: similar
        value: NYC
tip

Functions can use variables from test cases:

{
  type: "function",
  function: {
    description: "Get temperature in {{city}}"
    // ...
  }
}

They can also include functions that dynamically reference vars:

{
  type: "function",
  function: {
    name: "get_temperature",
    parameters: {
      type: "object",
        properties: {
          unit: {
            type: "string",
            enum: (vars) => vars.units,
          }
        },
    }
  }
}

Using functions​

functions and function_call is deprecated in favor of tools and tool_choice, see detail in OpenAI API reference.

Use the functions config to define custom functions. Each function should be an object with a name, optional description, and parameters. For example:

prompts: [prompt.txt]
providers:
  - id: openai:chat:gpt-4o-mini
    config:
      functions:
        [
          {
            'name': 'get_current_weather',
            'description': 'Get the current weather in a given location',
            'parameters':
              {
                'type': 'object',
                'properties':
                  {
                    'location':
                      {
                        'type': 'string',
                        'description': 'The city and state, e.g. San Francisco, CA',
                      },
                    'unit': { 'type': 'string', 'enum': ['celsius', 'fahrenheit'] },
                  },
                'required': ['location'],
              },
          },
        ]
tests:
  - vars:
      city: Boston
    assert:
      - type: is-valid-openai-function-call
  - vars:
      city: New York
  # ...

Sometimes OpenAI function calls don't match functions schemas. Use is-valid-openai-function-call assertions to enforce an exact schema match between function calls and the function definition.

To further test function call definitions, you can use the javascript assertion and/or transform directives. For example:

tests:
  - vars:
      city: Boston
    assert:
      - type: is-valid-openai-function-call
      - type: javascript
        value: output.name === 'get_current_weather'
      - type: javascript
        value: JSON.parse(output.arguments).location === 'Boston, MA'

  - vars:
      city: New York
    # transform returns only the 'name' property for this testcase
    transform: output.name
    assert:
      - type: is-json
      - type: similar
        value: NYC

Loading tools/functions from a file​

Instead of duplicating function definitions across multiple configurations, you can reference an external YAML (or JSON) file that contains your functions. This allows you to maintain a single source of truth for your functions, which is particularly useful if you have multiple versions or regular changes to definitions.

To load your functions from a file, specify the file path in your provider configuration like so:

providers:
  - file://./path/to/provider_with_function.yaml

Here's an example of how your provider_with_function.yaml might look:

id: openai:chat:gpt-4o-mini
config:
  functions:
    - name: get_current_weather
      description: Get the current weather in a given location
      parameters:
        type: object
        properties:
          location:
            type: string
            description: The city and state, e.g. San Francisco, CA
          unit:
            type: string
            enum:
              - celsius
              - fahrenheit
            description: The unit in which to return the temperature
        required:
          - location

Supported environment variables​

These OpenAI-related environment variables are supported:

VariableDescription
OPENAI_TEMPERATURETemperature model parameter, defaults to 0.
OPENAI_MAX_TOKENSMax_tokens model parameter, defaults to 1024.
OPENAI_API_HOSTThe hostname to use (useful if you're using an API proxy). Takes priority over OPENAI_BASE_URL.
OPENAI_BASE_URLThe base URL (protocol + hostname + port) to use, this is a more general option than OPENAI_API_HOST.
OPENAI_API_KEYOpenAI API key.
OPENAI_ORGANIZATIONThe OpenAI organization key to use.
PROMPTFOO_DELAY_MSNumber of milliseconds to delay between API calls. Useful if you are hitting OpenAI rate limits (defaults to 0).
PROMPTFOO_REQUEST_BACKOFF_MSBase number of milliseconds to backoff and retry if a request fails (defaults to 5000).

Evaluating assistants​

To test out an Assistant via OpenAI's Assistants API, first create an Assistant in the API playground.

Set functions, code interpreter, and files for retrieval as necessary.

Then, include the assistant in your config:

prompts:
  - 'Write a tweet about {{topic}}'
providers:
  - openai:assistant:asst_fEhNN3MClMamLfKLkIaoIpgZ
tests:
  - vars:
      topic: bananas
  # ...

Code interpreter, function calls, and retrievals will be included in the output alongside chat messages. Note that the evaluator creates a new thread for each eval.

The following properties can be overwritten in provider config:

  • model - OpenAI model to use
  • instructions - System prompt
  • tools - Enabled tools
  • thread.messages - A list of message objects that the thread is created with.
  • temperature - Temperature for the model
  • toolChoice - Controls whether the AI should use a tool
  • attachments - File attachments to include in messages - see Assistant v2 attachments

Here's an example of a more detailed config:

prompts:
  - 'Write a tweet about {{topic}}'
providers:
  - id: openai:assistant:asst_fEhNN3MClMamLfKLkIaoIpgZ
    config:
      model: gpt-4-1106-preview
      instructions: "You always speak like a pirate"
      temperature: 0.2
      toolChoice:
        type: file_search
      tools:
        - type: code_interpreter
        - type: file_search
      thread:
        messages:
          - role: user
            content: "Hello world"
          - role: assistant
            content: "Greetings from the high seas"
tests:
  - vars:
      topic: bananas
  # ...

Automatically handling function tool calls​

You can specify JavaScript callbacks that are automatically called to create the output of a function tool call.

This requires defining your config in a JavaScript file instead of YAML.

module.exports = /** @type {import('promptfoo').TestSuiteConfig} */ ({
  prompts: 'Please add the following numbers together: {{a}} and {{b}}',
  providers: [
    {
      id: 'openai:assistant:asst_fEhNN3MClMamLfKLkIaoIpgZ',
      config:
        /** @type {InstanceType<import('promptfoo')["providers"]["OpenAiAssistantProvider"]>["config"]} */ ({
          model: 'gpt-4-1106-preview',
          instructions: 'You can add two numbers together using the `addNumbers` tool',
          tools: [
            {
              type: 'function',
              function: {
                name: 'addNumbers',
                description: 'Add two numbers together',
                parameters: {
                  type: 'object',
                  properties: {
                    a: { type: 'number' },
                    b: { type: 'number' },
                  },
                  required: ['a', 'b'],
                },
              },
            },
          ],
          /**
           * Map of function tool names to function callback.
           */
          functionToolCallbacks: {
            // this function should accept a string, and return a string
            // or a `Promise<string>`.
            addNumbers: (parametersJsonString) => {
              const { a, b } = JSON.parse(parametersJsonString);
              return JSON.stringify(a + b);
            },
          },
        }),
    },
  ],
  tests: [
    {
      vars: { a: 5, b: 6 },
    },
  ],
});

Troubleshooting​

OpenAI rate limits​

There are a few things you can do if you encounter OpenAI rate limits (most commonly with GPT-4):

  1. Reduce concurrency to 1 by setting --max-concurrency 1 in the CLI, or by setting evaluateOptions.maxConcurrency in the config.
  2. Set a delay between requests by setting --delay 3000 (3000 ms) in the CLI, or by setting evaluateOptions.delay in the config, or with the environment variable PROMPTFOO_DELAY_MS (all values are in milliseconds).
  3. Adjust the exponential backoff for failed requests by setting the environment variable PROMPTFOO_REQUEST_BACKOFF_MS. This defaults to 5000 milliseconds and retries exponential up to 4 times. You can increase this value if requests are still failing, but note that this can significantly increase end-to-end test time.

OpenAI flakiness​

To retry HTTP requests that are Internal Server errors, set the PROMPTFOO_RETRY_5XX environment variable to 1.