Python SDK

Use the Python SDK to enforce guardrails in your applications. Automate safety checks at runtime.

1. Environment Setup

os.environ["RAGAAI_CATALYST_BASE_URL"] = "https://catalyst.raga.ai/api""
os.environ["OPENAI_API_KEY"] = "Your LLM API key"

• Configures the environment.

  • RAGAAI_CATALYST_BASE_URL: The base URL of the RagaAI Catalyst API for interaction.

  • OPENAI_API_KEY: Key to authenticate OpenAI-based LLM interactions.

---

2. Initializing the RagaAICatalyst Client

catalyst = RagaAICatalyst(
    access_key="Generate access key form settings",
    secret_key="Generate access key form settings",
)

• Creates an authenticated client to interact with RagaAI Catalyst.

  • access_key and secret_key: Credentials to access RagaAI Catalyst.

---

3. Initializing the Guardrails Manager

gdm = GuardrailsManager(project_name="Project Name")

• Purpose: Sets up a guardrails manager for managing guardrail-related configurations.

  • project_name: Links the guardrails to a specific project .

---

4. Listing Available Guardrails

guardrails_list = gdm.list_guardrails()
print('guardrails_list:', guardrails_list)

• Retrieves a list of all guardrails configured in the project.

• Output: A list of guardrails available for use.

---

5. Listing Fail Conditions

fail_conditions = gdm.list_fail_condition()
print('fail_conditions;', fail_conditions)

• Retrieves the conditions under which guardrails will flag a failure.

• Output: A list of fail conditions (e.g., ALL_FAIL, SOME_FAIL, etc.).

---

6. Retrieving Deployment IDs

pdeployment_list = gdm.list_deployment_ids()
print('deployment_list:', deployment_list)

• Lists all deployment IDs associated with guardrails.

• Output: A list of deployment IDs, each representing a configuration of guardrails.

---

7. Fetching Deployment Details

deployment_id_detail = gdm.get_deployment(17)
print('deployment_id_detail:', deployment_id_detail)

• Retrieves details of a specific deployment by its ID (17 in this example).

• Output: Details of the deployment, including associated guardrails and configurations.

---

8. Adding Guardrails to a Deployment

guardrails_config = {"guardrailFailConditions": ["FAIL"],
                     "deploymentFailCondition": "ALL_FAIL",
                     "alternateResponse": "Your alternate response"}

• Configures guardrail behavior when conditions fail.

  • guardrailFailConditions: Triggers guardrails when specific conditions are met.

  • deploymentFailCondition: Aggregates multiple failures (ALL_FAIL requires all guardrails to fail).

  • alternateResponse: Fallback response in case of guardrail-triggered failures.

guardrails = [
    {
      "displayName": "Response_Evaluator",
      "name": "Response Evaluator",
      "config": {
          "mappings": [{"schemaName": "Text", "variableName": "Response"}],
          "params": {
              "isActive": {"value": False},
              "isHighRisk": {"value": True},
              "threshold": {"eq": 0},
              "competitors": {"value": ["Google", "Amazon"]}
          }
      }
    },
    {
      "displayName": "Regex_Check",
      "name": "Regex Check",
      "config": {
          "mappings": [{"schemaName": "Text", "variableName": "Response"}],
          "params": {
              "isActive": {"value": False},
              "isHighRisk": {"value": True},
              "threshold": {"lt1": 1}
          }
      }
    }
]

• Purpose: Defines guardrail configurations.

  • displayName: A user-friendly name for the guardrail.

  • name: The internal name of the guardrail.

  • config: Contains mappings, parameters, and settings for the guardrail's logic.

---

9. Initializing the GuardExecutor

executor = GuardExecutor(17, gdm, field_map={'context': 'document'})

• Purpose: Initializes the executor to run evaluations using the deployment id.

  • 17: Deployment ID of the guardrails to apply.

  • field_map: Maps input fields (e.g., context) to expected variables (document).

---

10. Preparing Input for Evaluation

message = {'role': 'user', 'content': 'What is the capital of France?'}
prompt_params = {'document': 'Paris is not the capital of france'}
model_params = {'temperature': .7, 'model': 'gpt-4o-mini'}
llm_caller = 'litellm'

• Supplies input for evaluation.

  • message: Represents the user query.

  • prompt_params: Contextual data provided to the model.

  • model_params: Configuration for the LLM response generation.

  • llm_caller: Specifies the API or library to call the LLM.

---

11. Executing the Guardrails

executor([message], prompt_params, model_params, llm_caller)

• Runs the guardrails evaluation with the given inputs.