Skip to main content

Metering for AI Units

Your application needs to report business metrics that reflect the value of AI functionality and cover the cost of GenAI/LLM services. The generative AI hub can report these metrics to Unified Metering on your behalf.

Each LLM request must supply additional headers as documented in the SAP Help Portal.

Required headers:

  • X-USECASE-ID
  • X-BUSINESS-CONTEXT
  • X-LOCALTENANT-ID
  • X-PRODUCT-TYPE

Common Setup​

from gen_ai_hub.proxy import get_proxy_client
from gen_ai_hub.proxy.gen_ai_hub_proxy import temporary_headers_addition

METERING_HEADERS = {
    'X-USECASE-ID': 'my-usecase',
    'X-BUSINESS-CONTEXT': 'my-context',
    'X-LOCALTENANT-ID': 'tenant-123',
    'X-PRODUCT-TYPE': 'my-product'
}

proxy_client = get_proxy_client('gen-ai-hub')
proxy_client.set_headers_addition(headers=METERING_HEADERS)

Native LLM Clients​

Instance-level headers​

from gen_ai_hub.proxy.native.openai import OpenAI

# All requests from this client include metering headers
client = OpenAI(proxy_client=proxy_client)
response = client.chat.completions.create(
    model='gpt-4o',
    messages=[{'role': 'user', 'content': 'Hello!'}]
)
print(response.choices[0].message.content)

Request-level headers​

from gen_ai_hub.proxy.native.openai import OpenAI

client = OpenAI()

# Only this request includes metering headers
with temporary_headers_addition(headers=METERING_HEADERS):
    response = client.chat.completions.create(
        model='gpt-4o',
        messages=[{'role': 'user', 'content': 'Hello!'}]
    )
    print(response.choices[0].message.content)

Orchestration Service​

Instance-level headers​

from gen_ai_hub.orchestration_v2 import (
    OrchestrationService, OrchestrationConfig, ModuleConfig, Template,
    PromptTemplatingModuleConfig, LLMModelDetails, UserMessage
)

config = OrchestrationConfig(
    modules=ModuleConfig(
        prompt_templating=PromptTemplatingModuleConfig(
            prompt=Template(template=[UserMessage(content='{{?input}}')]),
            model=LLMModelDetails(name='gpt-4o')
        )
    )
)

# All requests from this service include metering headers
service = OrchestrationService(proxy_client=proxy_client, config=config)
response = service.run(placeholder_values={'input': 'Hello!'})
print(response.final_result.choices[0].message.content)

Request-level headers​

from gen_ai_hub.orchestration_v2 import OrchestrationService

service = OrchestrationService(config=config)

with temporary_headers_addition(headers=METERING_HEADERS):
    response = service.run(placeholder_values={'input': 'Hello!'})
    print(response.final_result.choices[0].message.content)

Prompt Registry​

Instance-level headers​

from gen_ai_hub.prompt_registry import PromptTemplateClient

client = PromptTemplateClient(proxy_client=proxy_client)
templates = client.get_prompt_templates(scenario='my-scenario')
print(f"Found {templates.count} templates")

Request-level headers​

from gen_ai_hub.prompt_registry import PromptTemplateClient

client = PromptTemplateClient()

with temporary_headers_addition(headers=METERING_HEADERS):
    templates = client.get_prompt_templates(scenario='my-scenario')
    print(f"Found {templates.count} templates")

Document Grounding Clients​

Instance-level headers​

from gen_ai_hub.document_grounding import PipelineAPIClient, RetrievalAPIClient, VectorAPIClient

pipeline_client = PipelineAPIClient(proxy_client=proxy_client)
retrieval_client = RetrievalAPIClient(proxy_client=proxy_client)
vector_client = VectorAPIClient(proxy_client=proxy_client)

print(f"Found {pipeline_client.get_pipelines().count} pipelines")
print(f"Found {retrieval_client.get_data_repositories().count} repositories")
print(f"Found {vector_client.get_collections().count} collections")

Request-level headers​

from gen_ai_hub.document_grounding import RetrievalAPIClient

client = RetrievalAPIClient()

with temporary_headers_addition(headers=METERING_HEADERS):
    repositories = client.get_data_repositories()
    print(f"Found {repositories.count} repositories")

Combining Instance and Request-level Headers​

Request-level headers are merged with instance-level headers. If the same header is set at both levels, the request-level value takes precedence.

from gen_ai_hub.proxy.native.openai import OpenAI

proxy_client.set_headers_addition({
    'X-USECASE-ID': 'default-usecase',
    'X-LOCALTENANT-ID': 'tenant-123'
})

client = OpenAI(proxy_client=proxy_client)

with temporary_headers_addition({'X-USECASE-ID': 'special-usecase'}):
    # X-USECASE-ID: 'special-usecase' (request-level wins)
    # X-LOCALTENANT-ID: 'tenant-123' (from instance-level)
    response = client.chat.completions.create(
        model='gpt-4o',
        messages=[{'role': 'user', 'content': 'Hello!'}]
    )