Skip to main content

openai-textgen

modify records using openai models

Description​

textgen is a conduit processor that will transform a record based on a given prompt

Configuration parameters​

version: 2.2
pipelines:
  - id: example
    status: running
    connectors:
      # define source and destination ...
    processors:
      - id: example
        plugin: "openai-textgen"
        settings:
          # APIKey is the OpenAI API key. Required.
          # Type: string
          api_key: ""
          # BackoffFactor is the factor by which the backoff increases. Defaults
          # to 2.0
          # Type: float
          backoff_factor: "2.0"
          # DeveloperMessage is the system message that guides the model's
          # behavior. Required.
          # Type: string
          developer_message: ""
          # Field is the reference to the field to process. Defaults to
          # ".Payload.After".
          # Type: string
          field: ".Payload.After"
          # FrequencyPenalty penalizes new tokens based on frequency in text.
          # Type: float
          frequency_penalty: ""
          # InitialBackoff is the initial backoff duration in milliseconds.
          # Defaults to 1000ms (1s).
          # Type: int
          initial_backoff: "1000"
          # LogProbs is whether to return log probabilities of output tokens.
          # Type: bool
          log_probs: ""
          # LogitBias modifies the likelihood of specified tokens appearing.
          # Type: int
          logit_bias.*: ""
          # MaxBackoff is the maximum backoff duration in milliseconds. Defaults
          # to 30000ms (30s).
          # Type: int
          max_backoff: "30000"
          # MaxCompletionTokens is the maximum number of tokens for completion.
          # Type: int
          max_completion_tokens: ""
          # MaxRetries is the maximum number of retries for API calls. Defaults
          # to 3.
          # Type: int
          max_retries: "3"
          # MaxTokens is the maximum number of tokens to generate.
          # Type: int
          max_tokens: ""
          # Metadata is additional metadata to include with the request.
          # Type: string
          metadata.*: ""
          # Model is the OpenAI model to use (e.g., gpt-4o-mini). Required.
          # Type: string
          model: ""
          # N is the number of completions to generate.
          # Type: int
          n: ""
          # PresencePenalty penalizes new tokens based on presence in text.
          # Type: float
          presence_penalty: ""
          # ReasoningEffort controls the amount of reasoning in the response.
          # Type: string
          reasoning_effort: ""
          # Whether to decode the record key using its corresponding schema from
          # the schema registry.
          # Type: bool
          sdk.schema.decode.key.enabled: "true"
          # Whether to decode the record payload using its corresponding schema
          # from the schema registry.
          # Type: bool
          sdk.schema.decode.payload.enabled: "true"
          # Whether to encode the record key using its corresponding schema from
          # the schema registry.
          # Type: bool
          sdk.schema.encode.key.enabled: "true"
          # Whether to encode the record payload using its corresponding schema
          # from the schema registry.
          # Type: bool
          sdk.schema.encode.payload.enabled: "true"
          # Seed is the seed for deterministic results.
          # Type: int
          seed: ""
          # Stop are sequences where the API will stop generating.
          # Type: string
          stop: ""
          # Store is whether to store the conversation in OpenAI.
          # Type: bool
          store: ""
          # Stream is whether to stream the results or not. Not used for now.
          # Type: bool
          stream: ""
          # StrictOutput enforces strict output format. Defaults to false.
          # Type: bool
          strict_output: "false"
          # Temperature controls randomness (0-2, lower is more deterministic).
          # Type: float
          temperature: ""
          # TopLogProbs is the number of most likely tokens to return
          # probabilities for.
          # Type: int
          top_log_probs: ""
          # TopP controls diversity via nucleus sampling.
          # Type: float
          top_p: ""
          # User is the user identifier for OpenAI API.
          # Type: string
          user: ""

Examples​

Transform text using OpenAI models​

This example shows how to use the OpenAI text generation processor to transform a record's .Payload.After field using an OpenAI model. The processor will send the content of the field to OpenAI and replace it with the response.

In this example, we're using a system message that instructs the model to convert the input text to uppercase.

Configuration parameters​

version: 2.2
pipelines:
  - id: example
    status: running
    connectors:
      # define source and destination ...
    processors:
      - id: example
        plugin: "openai-textgen"
        settings:
          api_key: "fake-api-key"
          backoff_factor: "2.0"
          developer_message: "You will receive a payload. Your task is to output back the payload in uppercase."
          field: ".Payload.After"
          initial_backoff: "1000"
          max_backoff: "30000"
          max_retries: "3"
          model: "gpt-4o-mini"
          strict_output: "false"
          temperature: "0"

Record difference​

Before
After
1
{
1
{
2
  "position": "cG9zLTE=",
2
  "position": "cG9zLTE=",
3
  "operation": "create",
3
  "operation": "create",
4
  "metadata": null,
4
  "metadata": null,
5
  "key": null,
5
  "key": null,
6
  "payload": {
6
  "payload": {
7
    "before": null,
7
    "before": null,
8
-
    "after": "hello world"
8
+
    "after": "HELLO WORLD"
9
  }
9
  }
10
}
10
}


scarf pixel conduit-site-docs-using-processors