revenium-python-sdk 0.14.0__tar.gz → 0.15.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/PKG-INFO +21 -10
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/README.md +19 -8
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/pyproject.toml +5 -2
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/__init__.py +4 -1
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/cache_tokens.py +101 -27
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/enforcement.py +502 -162
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/job_type_economics.py +20 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/middleware.py +42 -1
- revenium_python_sdk-0.15.0/revenium_middleware/litellm/proxy/_health_check.py +50 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/proxy/guardrail.py +5 -1
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/proxy/middleware.py +7 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_python_sdk.egg-info/PKG-INFO +21 -10
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_python_sdk.egg-info/SOURCES.txt +1 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_python_sdk.egg-info/requires.txt +1 -1
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/LICENSE +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/call_context.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/call_ownership.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/closable_scope.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/config.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/context.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/decorators.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/delivery_circuit.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/error_text.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/exceptions.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/failed_calls.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/json_safe.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/langchain_settings.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/load_diagnostics.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/log_sanitize.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/metering.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/metering_buffer.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/metering_pool.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/metering_status.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/metering_submission.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/outcomes.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/package_versions.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/parse_failures.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/patch_registry.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/perplexity_cost.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/perplexity_usage_metadata.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/prompt_extraction.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/responses_status.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/secret_shapes.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/selective_metering.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/shutdown_signals.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/sse_body_tap.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/subscriber.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/tool_event_transport.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/usage_iterations.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/vendor_model_aliases.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/wire_guards.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/LICENSE +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_base_client.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_client.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_compat.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_constants.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_exceptions.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_files.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_models.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_qs.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_resource.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_response.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_streaming.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_types.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_logs.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_proxy.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_reflection.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_resources_proxy.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_streams.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_sync.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_transform.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_typing.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_utils/_utils.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/_version.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/context.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/decorator.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/py.typed +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/resources/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/resources/ai.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/resources/apis.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/resources/events.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/ai_create_audio_params.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/ai_create_completion_params.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/ai_create_image_params.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/ai_create_video_params.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/api_meter_request_params.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/api_meter_response_params.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/event_create_params.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_metering/types/metering_response_resource.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/agentic_outcomes.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/bedrock_adapter.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/bedrock_model_names.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/bedrock_transport.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/config.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/inference_profiles.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/middleware.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/provider.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/raw_response.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/stream_create.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/anthropic/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/fal/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/fal/_call.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/fal/_metering.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/fal/_queue.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/fal/config.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/fal/middleware.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/fal/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/common/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/common/exceptions.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/common/protocols.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/common/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/common/types.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/common/utils.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/config.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/google_ai/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/google_ai/middleware.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/google_ai/provider.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/vertex_ai/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/vertex_ai/middleware.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/google/vertex_ai/provider.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/_metadata.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/_retry.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/anthropic_driver.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/budget_errors.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/litellm_driver.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/ollama_driver.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/openai_driver.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/openai_embedding_driver.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/griptape/universal_driver.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/job_context.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/job_history.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/config.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/context.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/decorators.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/hooks.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/integrations/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/integrations/crewai.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/client/validation.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/proxy/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/litellm/proxy/_metering_owner.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/ollama/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/ollama/middleware.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/ollama/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/audio.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/azure_config.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/azure_model_resolver.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/config.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/exceptions.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/_abandoned_runs.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/_model_call.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/_perplexity_cost_capture.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/_prompt_capture.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/_text_completion_usage.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/_utils.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/langchain/unified_handler.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/middleware.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/provider.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/raw_response.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/openai/xai_usage.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/_metering_scope.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/middleware.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/patching.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/perplexity_sdk.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/provider.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/raw_response.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/streaming.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/perplexity/trace_fields.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/webhooks/__init__.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/webhooks/_verify.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_python_sdk.egg-info/dependency_links.txt +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_python_sdk.egg-info/top_level.txt +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/setup.cfg +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/tests/test_metering.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/tests/test_metering_dispatch_timestamps.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/tests/test_patch_coverage.py +0 -0
- {revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/tests/test_release_verify_ticket.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: revenium-python-sdk
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.15.0
|
|
4
4
|
Summary: The official Revenium Python SDK — unified AI metering middleware for OpenAI, Anthropic, Google, Ollama, LiteLLM, Perplexity, and fal.ai.
|
|
5
5
|
Author-email: Revenium <support@revenium.io>
|
|
6
6
|
License: MIT
|
|
@@ -80,7 +80,7 @@ Requires-Dist: litellm>=1.40.0; extra == "litellm"
|
|
|
80
80
|
Requires-Dist: python-dotenv>=0.19.0; extra == "litellm"
|
|
81
81
|
Provides-Extra: litellm-proxy
|
|
82
82
|
Requires-Dist: wrapt>=1.14.0; extra == "litellm-proxy"
|
|
83
|
-
Requires-Dist: litellm[proxy]>=1.
|
|
83
|
+
Requires-Dist: litellm[proxy]>=1.94.0; extra == "litellm-proxy"
|
|
84
84
|
Provides-Extra: dev
|
|
85
85
|
Requires-Dist: pytest>=7.0.0; extra == "dev"
|
|
86
86
|
Requires-Dist: pytest-asyncio; extra == "dev"
|
|
@@ -196,11 +196,11 @@ The calls below are billed by the provider (some, such as batch and fine-tuning
|
|
|
196
196
|
<!-- not-metered:start -->
|
|
197
197
|
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.create`, `beta.agents.automations.create`, `beta.agents.automations.invocations.create`, `beta.agents.sessions.create`, `beta.agents.sessions.events.create`, `beta.agents.sessions.events.stream`, `beta.agents.sessions.stream`, `beta.agents.sessions.update`, `beta.chatkit.sessions.create`, `beta.realtime.connect`, `beta.responses.compact`, `beta.responses.connect`, `beta.responses.create`, `beta.threads.create_and_run`, `beta.threads.create_and_run_poll`, `beta.threads.create_and_run_stream`, `beta.threads.runs.create`, `beta.threads.runs.create_and_poll`, `beta.threads.runs.create_and_stream`, `beta.threads.runs.stream`, `beta.threads.runs.submit_tool_outputs`, `beta.threads.runs.submit_tool_outputs_and_poll`, `beta.threads.runs.submit_tool_outputs_stream`, `completions.create`, `decisions.create`, `evals.runs.create`, `fine_tuning.alpha.graders.run`, `fine_tuning.jobs.create`, `images.create_variation`, `images.edit`, `images.generate`, `live.connect`, `live.create`, `live.forks.connect`, `live.sessions.accept`, `live.sessions.fork`, `live.sideband.connect`, `realtime.calls.accept`, `realtime.calls.create`, `realtime.connect`, `realtime.translations.connect`, `responses.compact`, `responses.connect`, `videos.create`, `videos.create_and_poll`, `videos.create_character`, `videos.edit`, `videos.extend`, `videos.remix`
|
|
198
198
|
- **Anthropic** (`Anthropic` and `AsyncAnthropic` clients): `beta.deployments.run`, `beta.dreams.create`, `beta.environments.work.poller`, `beta.environments.work.worker`, `beta.messages.batches.create`, `beta.sessions.create`, `beta.sessions.events.send`, `beta.sessions.events.stream`, `beta.sessions.events.tool_runner`, `beta.sessions.threads.events.stream`, `messages.batches.create`
|
|
199
|
-
- **Google Gemini** (`google.genai` `Client` and `Client.aio`): `batches.create`, `batches.create_embeddings`, `caches.create`, `caches.update`, `file_search_stores.import_file`, `file_search_stores.upload_to_file_search_store`, `interactions.create`, `live.connect`, `live.music.connect`, `models.edit_image`, `models.recontext_image`, `models.segment_image`, `models.upscale_image`, `triggers.run`, `tunings.tune`
|
|
199
|
+
- **Google Gemini** (`google.genai` `Client` and `Client.aio`): `batches.create`, `batches.create_embeddings`, `caches.create`, `caches.update`, `decisions.create`, `file_search_stores.import_file`, `file_search_stores.upload_to_file_search_store`, `interactions.create`, `live.connect`, `live.music.connect`, `models.edit_image`, `models.recontext_image`, `models.segment_image`, `models.upscale_image`, `triggers.run`, `tunings.tune`
|
|
200
200
|
- **Google Vertex AI** (`vertexai` classes): `batch_prediction.BatchPredictionJob.submit`, `caching.CachedContent.create`, `caching.CachedContent.update`, `language_models.ChatModel.tune_model`, `language_models.ChatModel.tune_model_rlhf`, `language_models.ChatSession.send_message`, `language_models.ChatSession.send_message_async`, `language_models.ChatSession.send_message_streaming`, `language_models.ChatSession.send_message_streaming_async`, `language_models.CodeChatModel.tune_model`, `language_models.CodeChatSession.send_message`, `language_models.CodeChatSession.send_message_async`, `language_models.CodeChatSession.send_message_streaming`, `language_models.CodeChatSession.send_message_streaming_async`, `language_models.CodeGenerationModel.batch_predict`, `language_models.CodeGenerationModel.predict`, `language_models.CodeGenerationModel.predict_async`, `language_models.CodeGenerationModel.predict_streaming`, `language_models.CodeGenerationModel.predict_streaming_async`, `language_models.CodeGenerationModel.tune_model`, `language_models.TextEmbeddingModel.batch_predict`, `language_models.TextEmbeddingModel.deploy_tuned_model`, `language_models.TextEmbeddingModel.tune_model`, `language_models.TextGenerationModel.batch_predict`, `language_models.TextGenerationModel.predict`, `language_models.TextGenerationModel.predict_async`, `language_models.TextGenerationModel.predict_streaming`, `language_models.TextGenerationModel.predict_streaming_async`, `language_models.TextGenerationModel.tune_model`, `language_models.TextGenerationModel.tune_model_rlhf`, `preview.caching.CachedContent.create`, `preview.caching.CachedContent.update`, `preview.vision_models.ImageCaptioningModel.get_captions`, `preview.vision_models.ImageGenerationModel.upscale_image`, `preview.vision_models.ImageQnAModel.ask_question`, `preview.vision_models.ImageSegmentationModel.segment_image`, `preview.vision_models.ImageTextModel.ask_question`, `preview.vision_models.ImageTextModel.get_captions`, `preview.vision_models.MultiModalEmbeddingModel.get_embeddings`, `preview.vision_models.WatermarkVerificationModel.verify_image`, `tuning.sft.train`, `vision_models.ImageCaptioningModel.get_captions`, `vision_models.ImageGenerationModel.upscale_image`, `vision_models.ImageQnAModel.ask_question`, `vision_models.ImageTextModel.ask_question`, `vision_models.ImageTextModel.get_captions`, `vision_models.MultiModalEmbeddingModel.get_embeddings`
|
|
201
201
|
- **AWS Bedrock** (boto3 bedrock-runtime client; Claude models are metered when REVENIUM_BEDROCK_TRANSPORT is on): `apply_guardrail`, `converse`, `converse_stream`, `invoke_guardrail_checks`, `invoke_model`, `invoke_model_with_response_stream`, `start_async_invoke`
|
|
202
202
|
- **Perplexity** (`Perplexity` and `AsyncPerplexity` clients): `async_.chat.completions.create`, `batches.create`, `browser.sessions.create`, `contextualized_embeddings.create`, `decisions.create`, `embeddings.create`, `responses.create`, `search.create`
|
|
203
|
-
- **LiteLLM** (the `litellm` module): `aadapter_completion`, `aadapter_generate_content`, `acode_interpreter_tool`, `acompact_responses`, `acreate_batch`, `acreate_fine_tuning_job`, `acreate_run`, `acreate_sandbox`, `adapter_completion`, `adecisions`, `agenerate_content`, `aimage_edit`, `aimage_generation`, `aimage_variation`, `aingest`, `anthropic_interface.messages.acreate`, `anthropic_interface.messages.create`, `anthropic_messages`, `anthropic_messages_handler`, `aocr`, `aquery`, `arealtime_calls`, `arerank`, `aresponses`, `aresponses_api_with_mcp`, `aresponses_with_retries`, `arun_code`, `arun_thread`, `arun_thread_stream`, `asearch`, `aspeech`, `asystemone`, `atranscription`, `avideo_create_character`, `avideo_edit`, `avideo_extension`, `avideo_generation`, `avideo_remix`, `compact_responses`, `config_completion`, `create_batch`, `create_fine_tuning_job`, `create_run`, `decisions`, `image_edit`, `image_generation`, `image_variation`, `ingest`, `nlp_cloud_chat_completion`, `ocr`, `query`, `replicate_chat_completion`, `rerank`, `responses`, `responses_with_retries`, `run_thread`, `run_thread_stream`, `search`, `speech`, `systemone`, `transcription`, `video_create_character`, `video_edit`, `video_extension`, `video_generation`, `video_remix`
|
|
203
|
+
- **LiteLLM** (the `litellm` module): `aadapter_completion`, `aadapter_generate_content`, `acode_interpreter_tool`, `acompact_responses`, `acreate_batch`, `acreate_fine_tuning_job`, `acreate_run`, `acreate_sandbox`, `adapter_completion`, `adecisions`, `aextraction`, `agenerate_content`, `aimage_edit`, `aimage_generation`, `aimage_variation`, `aingest`, `anthropic_interface.messages.acreate`, `anthropic_interface.messages.create`, `anthropic_messages`, `anthropic_messages_handler`, `aocr`, `aquery`, `arealtime_calls`, `arerank`, `aresponses`, `aresponses_api_with_mcp`, `aresponses_with_retries`, `arun_code`, `arun_thread`, `arun_thread_stream`, `asearch`, `aspeech`, `asystemone`, `atranscription`, `avideo_create_character`, `avideo_edit`, `avideo_extension`, `avideo_generation`, `avideo_remix`, `compact_responses`, `config_completion`, `create_batch`, `create_fine_tuning_job`, `create_run`, `decisions`, `extraction`, `image_edit`, `image_generation`, `image_variation`, `ingest`, `nlp_cloud_chat_completion`, `ocr`, `query`, `replicate_chat_completion`, `rerank`, `responses`, `responses_with_retries`, `run_thread`, `run_thread_stream`, `search`, `speech`, `systemone`, `transcription`, `video_create_character`, `video_edit`, `video_extension`, `video_generation`, `video_remix`
|
|
204
204
|
- **Ollama** (`Client`, `AsyncClient` and the `ollama` module): `systemone`, `web_fetch`, `web_search`
|
|
205
205
|
- **fal.ai** (`SyncClient`, `AsyncClient` and the `fal_client` module): `realtime`, `realtime_async`, `ws_connect`, `ws_connect_async`
|
|
206
206
|
- **LangChain** (embeddings models from vendors other than OpenAI, Azure OpenAI and Ollama, such as `MistralAIEmbeddings`): `Embeddings.aembed_documents`, `Embeddings.aembed_query`, `Embeddings.embed_documents`, `Embeddings.embed_query`
|
|
@@ -382,6 +382,13 @@ explicit override. Economics metric directions are
|
|
|
382
382
|
Use `CUSTOMER_DECLARED` or `MEASURED` for a baseline override. Use `MEASURED`,
|
|
383
383
|
`SELF_REPORTED`, or `DERIVED` for a period fact override.
|
|
384
384
|
|
|
385
|
+
A baseline can also record the work's throughput and headcount before AI:
|
|
386
|
+
`throughput_value` with a `throughput_period` of `HOUR`, `DAY`, `WEEK`, or
|
|
387
|
+
`MONTH`, an optional `throughput_scope` of `TEAM` (the default) or
|
|
388
|
+
`PER_PERSON`, and `headcount`. Send `throughput_period` whenever
|
|
389
|
+
`throughput_value` is set; the server rejects a period or scope without a
|
|
390
|
+
value.
|
|
391
|
+
|
|
385
392
|
Facts are append-only and keyed on the period, dimension and metric key
|
|
386
393
|
together. Re-appending that tuple supersedes the active fact, and the server
|
|
387
394
|
requires `reason=` on the entry when it does.
|
|
@@ -1188,6 +1195,10 @@ except calls a spending cap refuses.
|
|
|
1188
1195
|
pip install "revenium-python-sdk[litellm-proxy]" # requires Python 3.10+
|
|
1189
1196
|
```
|
|
1190
1197
|
|
|
1198
|
+
The extra requires **LiteLLM 1.94.0 or newer**. On LiteLLM 1.93.x the proxy cannot
|
|
1199
|
+
load the guardrail from the config below: it only looks for a module file beside
|
|
1200
|
+
`config.yaml` and stops with `Could not find module file`.
|
|
1201
|
+
|
|
1191
1202
|
```yaml
|
|
1192
1203
|
guardrails:
|
|
1193
1204
|
- guardrail_name: "revenium"
|
|
@@ -1235,7 +1246,7 @@ the call proceeds, and it proceeds without waiting. The pre-call check reads the
|
|
|
1235
1246
|
rules the SDK already has cached and refreshes them on a background thread, so a
|
|
1236
1247
|
Revenium endpoint that hangs, refuses connections, returns errors or rejects your
|
|
1237
1248
|
key (403) neither blocks nor slows the proxy. A failed refresh is retried once
|
|
1238
|
-
per `REVENIUM_CB_POLL_INTERVAL_SECONDS
|
|
1249
|
+
per `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s), not once per request. Right after the
|
|
1239
1250
|
proxy starts, before the first rules fetch completes, no rules are cached and
|
|
1240
1251
|
calls go through; see [Fail-Open vs Fail-Closed](#fail-open-vs-fail-closed).
|
|
1241
1252
|
Metering is likewise non-disruptive — nothing in the post-call path can turn a
|
|
@@ -1433,7 +1444,7 @@ Blank or unset means on; `true`, `1`, `yes`, `on` mean on; any other value turns
|
|
|
1433
1444
|
New calls go straight back to two records when you do; records already merged stay
|
|
1434
1445
|
merged. Four things have to be true for it to work:
|
|
1435
1446
|
|
|
1436
|
-
* **LiteLLM 1.
|
|
1447
|
+
* **LiteLLM 1.94.0 or newer.** That is the floor of the `litellm-proxy` extra, which
|
|
1437
1448
|
is what installs the guardrail. On an older LiteLLM the guardrail logs one warning
|
|
1438
1449
|
at startup and behaves exactly as it does when opted out: no header is added,
|
|
1439
1450
|
the provider's response id is reported, and you get two records rather than none.
|
|
@@ -2390,7 +2401,7 @@ REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional:
|
|
|
2390
2401
|
| `REVENIUM_BYPASS` | `false` | When `true`, every `check_enforcement` call short-circuits to a no-op. Useful for incident response. |
|
|
2391
2402
|
| `REVENIUM_TEAM_ID` | — | Hashed team ID. Path component on rule fetches; required when the breaker is enabled. |
|
|
2392
2403
|
| `REVENIUM_ENFORCEMENT_BASE_URL` | `https://api.revenium.ai/profitstream` | Base URL for the enforcement API, taken as given. When unset it is the address outcomes and jobs use (`REVENIUM_PROFITSTREAM_BASE_URL`, else the Revenium address `REVENIUM_METERING_BASE_URL` or `initialize_metering(base_url=...)` sits under, without its `/meter` segment) plus `/profitstream`. A rules fetch that answers 404 is retried once at the same address with `/profitstream` removed, or added if it was missing, and the address that answers with rules is used from then on, so a self-hosted server that serves the API at the root, or a value set without `/profitstream`, still reaches the rules. |
|
|
2393
|
-
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `
|
|
2404
|
+
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `30` | Seconds between background rule refreshes, and the wait before a failed refresh is retried. Whole seconds from a minimum of `5` to a maximum of `3600`: a lower value is raised to `5` with one warning per process start, and a higher one is capped at `3600`. Anything other than plain digits after trimming spaces (a sign, a decimal point, other text), an empty value and `0` use the default. Read once when polling starts, so a change takes effect when the process restarts. While calls are arriving, rules older than 120 seconds are refreshed on the next call whatever the interval, so a value above `120` only spaces out refreshes on an idle process and the retry after a failed refresh. |
|
|
2394
2405
|
| `REVENIUM_CB_FAIL_MODE` | `open` | `open` (default) lets calls through when no cache exists; `closed` raises `BudgetExceededError` until rules are loaded. |
|
|
2395
2406
|
| `REVENIUM_CACHE_DIR` | unset | When set, the rule cache is mirrored to `<dir>/revenium_enforcement_rules.<scope>.json` so a restarted process doesn't fail-closed on the very first call. `<scope>` is a short hash of the enforcement base URL and `REVENIUM_TEAM_ID`, so services for different teams can share one folder and each loads only its own rules. Nothing is saved without `REVENIUM_TEAM_ID`. |
|
|
2396
2407
|
|
|
@@ -2423,9 +2434,9 @@ LangChain embeddings models run no callbacks: `OpenAIEmbeddings` and `AzureOpenA
|
|
|
2423
2434
|
When the circuit breaker is disabled, it is a no-op. When enabled:
|
|
2424
2435
|
|
|
2425
2436
|
1. A daemon thread (`revenium-enforcement-poll`) starts on first use.
|
|
2426
|
-
2. It polls `GET {REVENIUM_ENFORCEMENT_BASE_URL}/v2/api/ai/enforcement-rules/{REVENIUM_TEAM_ID}` every `REVENIUM_CB_POLL_INTERVAL_SECONDS` with the `x-api-key` header.
|
|
2437
|
+
2. It polls `GET {REVENIUM_ENFORCEMENT_BASE_URL}/v2/api/ai/enforcement-rules/{REVENIUM_TEAM_ID}` every `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s) with the `x-api-key` header.
|
|
2427
2438
|
3. Rules are cached in-process. The pre-call check only reads that cache: it never makes a network request and never sleeps, so it is safe on an event loop (LiteLLM proxy, `AsyncOpenAI`). When the cache is more than 120 s old the check wakes the poller to refresh it early and carries on with the rules it has. Only one refresh runs at a time per process.
|
|
2428
|
-
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS
|
|
2439
|
+
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s), however much traffic arrives meanwhile. A 4xx answer the SDK does not retry (any but 408 and 429) is logged at WARNING with the URL and the status, so a wrong address (404) or a wrong key or team id (401, 403) is visible without debug logging.
|
|
2429
2440
|
5. `204 No Content` is treated as "no rules configured" — the cache is cleared.
|
|
2430
2441
|
|
|
2431
2442
|
A rule with filters (a cap scoped to an agent, model, provider, subscriber, credential, task type, product or organization) blocks or warns only the calls that match all of them; a filter the SDK cannot evaluate, such as the model of a LiteLLM proxy request, keeps the rule applying.
|
|
@@ -2533,7 +2544,7 @@ for budget in budgets or []:
|
|
|
2533
2544
|
|
|
2534
2545
|
Each row has `source` `ANOMALY` and `action` `NOTIFY`, groups by `SUBSCRIBER` with a per-person `groupBreakdown`, and carries `readingAt` when the current window has a reading. The call returns `[]` when the team has no advisory budgets and `None` when the enforcement API cannot be reached.
|
|
2535
2546
|
|
|
2536
|
-
The background poller is unaffected: it keeps reading the **whole team's** rules every `REVENIUM_CB_POLL_INTERVAL_SECONDS
|
|
2547
|
+
The background poller is unaffected: it keeps reading the **whole team's** rules every `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s). That is deliberate. The server computes the department-budget maps team-wide and attaches them to the team-wide read, so a poll narrowed to a single rule would stop receiving them and department budgets would quietly stop blocking anyone. Narrowing is an explicit, opt-in inspection call and never the refresh.
|
|
2537
2548
|
|
|
2538
2549
|
### End-to-End Example
|
|
2539
2550
|
|
|
@@ -80,11 +80,11 @@ The calls below are billed by the provider (some, such as batch and fine-tuning
|
|
|
80
80
|
<!-- not-metered:start -->
|
|
81
81
|
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.create`, `beta.agents.automations.create`, `beta.agents.automations.invocations.create`, `beta.agents.sessions.create`, `beta.agents.sessions.events.create`, `beta.agents.sessions.events.stream`, `beta.agents.sessions.stream`, `beta.agents.sessions.update`, `beta.chatkit.sessions.create`, `beta.realtime.connect`, `beta.responses.compact`, `beta.responses.connect`, `beta.responses.create`, `beta.threads.create_and_run`, `beta.threads.create_and_run_poll`, `beta.threads.create_and_run_stream`, `beta.threads.runs.create`, `beta.threads.runs.create_and_poll`, `beta.threads.runs.create_and_stream`, `beta.threads.runs.stream`, `beta.threads.runs.submit_tool_outputs`, `beta.threads.runs.submit_tool_outputs_and_poll`, `beta.threads.runs.submit_tool_outputs_stream`, `completions.create`, `decisions.create`, `evals.runs.create`, `fine_tuning.alpha.graders.run`, `fine_tuning.jobs.create`, `images.create_variation`, `images.edit`, `images.generate`, `live.connect`, `live.create`, `live.forks.connect`, `live.sessions.accept`, `live.sessions.fork`, `live.sideband.connect`, `realtime.calls.accept`, `realtime.calls.create`, `realtime.connect`, `realtime.translations.connect`, `responses.compact`, `responses.connect`, `videos.create`, `videos.create_and_poll`, `videos.create_character`, `videos.edit`, `videos.extend`, `videos.remix`
|
|
82
82
|
- **Anthropic** (`Anthropic` and `AsyncAnthropic` clients): `beta.deployments.run`, `beta.dreams.create`, `beta.environments.work.poller`, `beta.environments.work.worker`, `beta.messages.batches.create`, `beta.sessions.create`, `beta.sessions.events.send`, `beta.sessions.events.stream`, `beta.sessions.events.tool_runner`, `beta.sessions.threads.events.stream`, `messages.batches.create`
|
|
83
|
-
- **Google Gemini** (`google.genai` `Client` and `Client.aio`): `batches.create`, `batches.create_embeddings`, `caches.create`, `caches.update`, `file_search_stores.import_file`, `file_search_stores.upload_to_file_search_store`, `interactions.create`, `live.connect`, `live.music.connect`, `models.edit_image`, `models.recontext_image`, `models.segment_image`, `models.upscale_image`, `triggers.run`, `tunings.tune`
|
|
83
|
+
- **Google Gemini** (`google.genai` `Client` and `Client.aio`): `batches.create`, `batches.create_embeddings`, `caches.create`, `caches.update`, `decisions.create`, `file_search_stores.import_file`, `file_search_stores.upload_to_file_search_store`, `interactions.create`, `live.connect`, `live.music.connect`, `models.edit_image`, `models.recontext_image`, `models.segment_image`, `models.upscale_image`, `triggers.run`, `tunings.tune`
|
|
84
84
|
- **Google Vertex AI** (`vertexai` classes): `batch_prediction.BatchPredictionJob.submit`, `caching.CachedContent.create`, `caching.CachedContent.update`, `language_models.ChatModel.tune_model`, `language_models.ChatModel.tune_model_rlhf`, `language_models.ChatSession.send_message`, `language_models.ChatSession.send_message_async`, `language_models.ChatSession.send_message_streaming`, `language_models.ChatSession.send_message_streaming_async`, `language_models.CodeChatModel.tune_model`, `language_models.CodeChatSession.send_message`, `language_models.CodeChatSession.send_message_async`, `language_models.CodeChatSession.send_message_streaming`, `language_models.CodeChatSession.send_message_streaming_async`, `language_models.CodeGenerationModel.batch_predict`, `language_models.CodeGenerationModel.predict`, `language_models.CodeGenerationModel.predict_async`, `language_models.CodeGenerationModel.predict_streaming`, `language_models.CodeGenerationModel.predict_streaming_async`, `language_models.CodeGenerationModel.tune_model`, `language_models.TextEmbeddingModel.batch_predict`, `language_models.TextEmbeddingModel.deploy_tuned_model`, `language_models.TextEmbeddingModel.tune_model`, `language_models.TextGenerationModel.batch_predict`, `language_models.TextGenerationModel.predict`, `language_models.TextGenerationModel.predict_async`, `language_models.TextGenerationModel.predict_streaming`, `language_models.TextGenerationModel.predict_streaming_async`, `language_models.TextGenerationModel.tune_model`, `language_models.TextGenerationModel.tune_model_rlhf`, `preview.caching.CachedContent.create`, `preview.caching.CachedContent.update`, `preview.vision_models.ImageCaptioningModel.get_captions`, `preview.vision_models.ImageGenerationModel.upscale_image`, `preview.vision_models.ImageQnAModel.ask_question`, `preview.vision_models.ImageSegmentationModel.segment_image`, `preview.vision_models.ImageTextModel.ask_question`, `preview.vision_models.ImageTextModel.get_captions`, `preview.vision_models.MultiModalEmbeddingModel.get_embeddings`, `preview.vision_models.WatermarkVerificationModel.verify_image`, `tuning.sft.train`, `vision_models.ImageCaptioningModel.get_captions`, `vision_models.ImageGenerationModel.upscale_image`, `vision_models.ImageQnAModel.ask_question`, `vision_models.ImageTextModel.ask_question`, `vision_models.ImageTextModel.get_captions`, `vision_models.MultiModalEmbeddingModel.get_embeddings`
|
|
85
85
|
- **AWS Bedrock** (boto3 bedrock-runtime client; Claude models are metered when REVENIUM_BEDROCK_TRANSPORT is on): `apply_guardrail`, `converse`, `converse_stream`, `invoke_guardrail_checks`, `invoke_model`, `invoke_model_with_response_stream`, `start_async_invoke`
|
|
86
86
|
- **Perplexity** (`Perplexity` and `AsyncPerplexity` clients): `async_.chat.completions.create`, `batches.create`, `browser.sessions.create`, `contextualized_embeddings.create`, `decisions.create`, `embeddings.create`, `responses.create`, `search.create`
|
|
87
|
-
- **LiteLLM** (the `litellm` module): `aadapter_completion`, `aadapter_generate_content`, `acode_interpreter_tool`, `acompact_responses`, `acreate_batch`, `acreate_fine_tuning_job`, `acreate_run`, `acreate_sandbox`, `adapter_completion`, `adecisions`, `agenerate_content`, `aimage_edit`, `aimage_generation`, `aimage_variation`, `aingest`, `anthropic_interface.messages.acreate`, `anthropic_interface.messages.create`, `anthropic_messages`, `anthropic_messages_handler`, `aocr`, `aquery`, `arealtime_calls`, `arerank`, `aresponses`, `aresponses_api_with_mcp`, `aresponses_with_retries`, `arun_code`, `arun_thread`, `arun_thread_stream`, `asearch`, `aspeech`, `asystemone`, `atranscription`, `avideo_create_character`, `avideo_edit`, `avideo_extension`, `avideo_generation`, `avideo_remix`, `compact_responses`, `config_completion`, `create_batch`, `create_fine_tuning_job`, `create_run`, `decisions`, `image_edit`, `image_generation`, `image_variation`, `ingest`, `nlp_cloud_chat_completion`, `ocr`, `query`, `replicate_chat_completion`, `rerank`, `responses`, `responses_with_retries`, `run_thread`, `run_thread_stream`, `search`, `speech`, `systemone`, `transcription`, `video_create_character`, `video_edit`, `video_extension`, `video_generation`, `video_remix`
|
|
87
|
+
- **LiteLLM** (the `litellm` module): `aadapter_completion`, `aadapter_generate_content`, `acode_interpreter_tool`, `acompact_responses`, `acreate_batch`, `acreate_fine_tuning_job`, `acreate_run`, `acreate_sandbox`, `adapter_completion`, `adecisions`, `aextraction`, `agenerate_content`, `aimage_edit`, `aimage_generation`, `aimage_variation`, `aingest`, `anthropic_interface.messages.acreate`, `anthropic_interface.messages.create`, `anthropic_messages`, `anthropic_messages_handler`, `aocr`, `aquery`, `arealtime_calls`, `arerank`, `aresponses`, `aresponses_api_with_mcp`, `aresponses_with_retries`, `arun_code`, `arun_thread`, `arun_thread_stream`, `asearch`, `aspeech`, `asystemone`, `atranscription`, `avideo_create_character`, `avideo_edit`, `avideo_extension`, `avideo_generation`, `avideo_remix`, `compact_responses`, `config_completion`, `create_batch`, `create_fine_tuning_job`, `create_run`, `decisions`, `extraction`, `image_edit`, `image_generation`, `image_variation`, `ingest`, `nlp_cloud_chat_completion`, `ocr`, `query`, `replicate_chat_completion`, `rerank`, `responses`, `responses_with_retries`, `run_thread`, `run_thread_stream`, `search`, `speech`, `systemone`, `transcription`, `video_create_character`, `video_edit`, `video_extension`, `video_generation`, `video_remix`
|
|
88
88
|
- **Ollama** (`Client`, `AsyncClient` and the `ollama` module): `systemone`, `web_fetch`, `web_search`
|
|
89
89
|
- **fal.ai** (`SyncClient`, `AsyncClient` and the `fal_client` module): `realtime`, `realtime_async`, `ws_connect`, `ws_connect_async`
|
|
90
90
|
- **LangChain** (embeddings models from vendors other than OpenAI, Azure OpenAI and Ollama, such as `MistralAIEmbeddings`): `Embeddings.aembed_documents`, `Embeddings.aembed_query`, `Embeddings.embed_documents`, `Embeddings.embed_query`
|
|
@@ -266,6 +266,13 @@ explicit override. Economics metric directions are
|
|
|
266
266
|
Use `CUSTOMER_DECLARED` or `MEASURED` for a baseline override. Use `MEASURED`,
|
|
267
267
|
`SELF_REPORTED`, or `DERIVED` for a period fact override.
|
|
268
268
|
|
|
269
|
+
A baseline can also record the work's throughput and headcount before AI:
|
|
270
|
+
`throughput_value` with a `throughput_period` of `HOUR`, `DAY`, `WEEK`, or
|
|
271
|
+
`MONTH`, an optional `throughput_scope` of `TEAM` (the default) or
|
|
272
|
+
`PER_PERSON`, and `headcount`. Send `throughput_period` whenever
|
|
273
|
+
`throughput_value` is set; the server rejects a period or scope without a
|
|
274
|
+
value.
|
|
275
|
+
|
|
269
276
|
Facts are append-only and keyed on the period, dimension and metric key
|
|
270
277
|
together. Re-appending that tuple supersedes the active fact, and the server
|
|
271
278
|
requires `reason=` on the entry when it does.
|
|
@@ -1072,6 +1079,10 @@ except calls a spending cap refuses.
|
|
|
1072
1079
|
pip install "revenium-python-sdk[litellm-proxy]" # requires Python 3.10+
|
|
1073
1080
|
```
|
|
1074
1081
|
|
|
1082
|
+
The extra requires **LiteLLM 1.94.0 or newer**. On LiteLLM 1.93.x the proxy cannot
|
|
1083
|
+
load the guardrail from the config below: it only looks for a module file beside
|
|
1084
|
+
`config.yaml` and stops with `Could not find module file`.
|
|
1085
|
+
|
|
1075
1086
|
```yaml
|
|
1076
1087
|
guardrails:
|
|
1077
1088
|
- guardrail_name: "revenium"
|
|
@@ -1119,7 +1130,7 @@ the call proceeds, and it proceeds without waiting. The pre-call check reads the
|
|
|
1119
1130
|
rules the SDK already has cached and refreshes them on a background thread, so a
|
|
1120
1131
|
Revenium endpoint that hangs, refuses connections, returns errors or rejects your
|
|
1121
1132
|
key (403) neither blocks nor slows the proxy. A failed refresh is retried once
|
|
1122
|
-
per `REVENIUM_CB_POLL_INTERVAL_SECONDS
|
|
1133
|
+
per `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s), not once per request. Right after the
|
|
1123
1134
|
proxy starts, before the first rules fetch completes, no rules are cached and
|
|
1124
1135
|
calls go through; see [Fail-Open vs Fail-Closed](#fail-open-vs-fail-closed).
|
|
1125
1136
|
Metering is likewise non-disruptive — nothing in the post-call path can turn a
|
|
@@ -1317,7 +1328,7 @@ Blank or unset means on; `true`, `1`, `yes`, `on` mean on; any other value turns
|
|
|
1317
1328
|
New calls go straight back to two records when you do; records already merged stay
|
|
1318
1329
|
merged. Four things have to be true for it to work:
|
|
1319
1330
|
|
|
1320
|
-
* **LiteLLM 1.
|
|
1331
|
+
* **LiteLLM 1.94.0 or newer.** That is the floor of the `litellm-proxy` extra, which
|
|
1321
1332
|
is what installs the guardrail. On an older LiteLLM the guardrail logs one warning
|
|
1322
1333
|
at startup and behaves exactly as it does when opted out: no header is added,
|
|
1323
1334
|
the provider's response id is reported, and you get two records rather than none.
|
|
@@ -2274,7 +2285,7 @@ REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional:
|
|
|
2274
2285
|
| `REVENIUM_BYPASS` | `false` | When `true`, every `check_enforcement` call short-circuits to a no-op. Useful for incident response. |
|
|
2275
2286
|
| `REVENIUM_TEAM_ID` | — | Hashed team ID. Path component on rule fetches; required when the breaker is enabled. |
|
|
2276
2287
|
| `REVENIUM_ENFORCEMENT_BASE_URL` | `https://api.revenium.ai/profitstream` | Base URL for the enforcement API, taken as given. When unset it is the address outcomes and jobs use (`REVENIUM_PROFITSTREAM_BASE_URL`, else the Revenium address `REVENIUM_METERING_BASE_URL` or `initialize_metering(base_url=...)` sits under, without its `/meter` segment) plus `/profitstream`. A rules fetch that answers 404 is retried once at the same address with `/profitstream` removed, or added if it was missing, and the address that answers with rules is used from then on, so a self-hosted server that serves the API at the root, or a value set without `/profitstream`, still reaches the rules. |
|
|
2277
|
-
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `
|
|
2288
|
+
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `30` | Seconds between background rule refreshes, and the wait before a failed refresh is retried. Whole seconds from a minimum of `5` to a maximum of `3600`: a lower value is raised to `5` with one warning per process start, and a higher one is capped at `3600`. Anything other than plain digits after trimming spaces (a sign, a decimal point, other text), an empty value and `0` use the default. Read once when polling starts, so a change takes effect when the process restarts. While calls are arriving, rules older than 120 seconds are refreshed on the next call whatever the interval, so a value above `120` only spaces out refreshes on an idle process and the retry after a failed refresh. |
|
|
2278
2289
|
| `REVENIUM_CB_FAIL_MODE` | `open` | `open` (default) lets calls through when no cache exists; `closed` raises `BudgetExceededError` until rules are loaded. |
|
|
2279
2290
|
| `REVENIUM_CACHE_DIR` | unset | When set, the rule cache is mirrored to `<dir>/revenium_enforcement_rules.<scope>.json` so a restarted process doesn't fail-closed on the very first call. `<scope>` is a short hash of the enforcement base URL and `REVENIUM_TEAM_ID`, so services for different teams can share one folder and each loads only its own rules. Nothing is saved without `REVENIUM_TEAM_ID`. |
|
|
2280
2291
|
|
|
@@ -2307,9 +2318,9 @@ LangChain embeddings models run no callbacks: `OpenAIEmbeddings` and `AzureOpenA
|
|
|
2307
2318
|
When the circuit breaker is disabled, it is a no-op. When enabled:
|
|
2308
2319
|
|
|
2309
2320
|
1. A daemon thread (`revenium-enforcement-poll`) starts on first use.
|
|
2310
|
-
2. It polls `GET {REVENIUM_ENFORCEMENT_BASE_URL}/v2/api/ai/enforcement-rules/{REVENIUM_TEAM_ID}` every `REVENIUM_CB_POLL_INTERVAL_SECONDS` with the `x-api-key` header.
|
|
2321
|
+
2. It polls `GET {REVENIUM_ENFORCEMENT_BASE_URL}/v2/api/ai/enforcement-rules/{REVENIUM_TEAM_ID}` every `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s) with the `x-api-key` header.
|
|
2311
2322
|
3. Rules are cached in-process. The pre-call check only reads that cache: it never makes a network request and never sleeps, so it is safe on an event loop (LiteLLM proxy, `AsyncOpenAI`). When the cache is more than 120 s old the check wakes the poller to refresh it early and carries on with the rules it has. Only one refresh runs at a time per process.
|
|
2312
|
-
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS
|
|
2323
|
+
4. A refresh that fails (timeout, refused connection, 5xx, or any 4xx such as a 403 for a wrong key or team id) keeps the cached rules, logs a warning, and is not retried for one `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s), however much traffic arrives meanwhile. A 4xx answer the SDK does not retry (any but 408 and 429) is logged at WARNING with the URL and the status, so a wrong address (404) or a wrong key or team id (401, 403) is visible without debug logging.
|
|
2313
2324
|
5. `204 No Content` is treated as "no rules configured" — the cache is cleared.
|
|
2314
2325
|
|
|
2315
2326
|
A rule with filters (a cap scoped to an agent, model, provider, subscriber, credential, task type, product or organization) blocks or warns only the calls that match all of them; a filter the SDK cannot evaluate, such as the model of a LiteLLM proxy request, keeps the rule applying.
|
|
@@ -2417,7 +2428,7 @@ for budget in budgets or []:
|
|
|
2417
2428
|
|
|
2418
2429
|
Each row has `source` `ANOMALY` and `action` `NOTIFY`, groups by `SUBSCRIBER` with a per-person `groupBreakdown`, and carries `readingAt` when the current window has a reading. The call returns `[]` when the team has no advisory budgets and `None` when the enforcement API cannot be reached.
|
|
2419
2430
|
|
|
2420
|
-
The background poller is unaffected: it keeps reading the **whole team's** rules every `REVENIUM_CB_POLL_INTERVAL_SECONDS
|
|
2431
|
+
The background poller is unaffected: it keeps reading the **whole team's** rules every `REVENIUM_CB_POLL_INTERVAL_SECONDS` (default 30 s, minimum 5 s). That is deliberate. The server computes the department-budget maps team-wide and attaches them to the team-wide read, so a poll narrowed to a single rule would stop receiving them and department budgets would quietly stop blocking anyone. Narrowing is an explicit, opt-in inspection call and never the refresh.
|
|
2421
2432
|
|
|
2422
2433
|
### End-to-End Example
|
|
2423
2434
|
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "revenium-python-sdk"
|
|
7
|
-
version = "0.
|
|
7
|
+
version = "0.15.0"
|
|
8
8
|
description = "The official Revenium Python SDK — unified AI metering middleware for OpenAI, Anthropic, Google, Ollama, LiteLLM, Perplexity, and fal.ai."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
requires-python = ">=3.8"
|
|
@@ -110,9 +110,12 @@ litellm = [
|
|
|
110
110
|
# dispatch that meters streamed calls. litellm[proxy] itself requires Python
|
|
111
111
|
# 3.10+, which is why revenium_middleware.litellm.proxy.guardrail carries an
|
|
112
112
|
# import guard for it while the rest of the SDK stays on requires-python >=3.8.
|
|
113
|
+
# 1.94.0 is the first LiteLLM whose config loader (get_instance_fn) imports an
|
|
114
|
+
# installed package: 1.93.x only loads a module file sitting beside config.yaml,
|
|
115
|
+
# so the README's guardrail entry cannot load there.
|
|
113
116
|
litellm-proxy = [
|
|
114
117
|
"wrapt>=1.14.0",
|
|
115
|
-
"litellm[proxy]>=1.
|
|
118
|
+
"litellm[proxy]>=1.94.0",
|
|
116
119
|
]
|
|
117
120
|
|
|
118
121
|
dev = [
|
|
@@ -213,7 +213,8 @@ from .job_history import ( # noqa: E402
|
|
|
213
213
|
get_job_transactions, get_job_types, get_outcome_history, list_jobs,
|
|
214
214
|
)
|
|
215
215
|
from .job_type_economics import ( # noqa: E402
|
|
216
|
-
Baseline, BaselineProvenance,
|
|
216
|
+
Baseline, BaselineProvenance, BaselineThroughputPeriod,
|
|
217
|
+
BaselineThroughputScope, JobTypeEconomics, JobTypeMetricDirection,
|
|
217
218
|
JobTypeMonetizationBasis, JobTypeMonetizationCategory,
|
|
218
219
|
OutcomeMetricProvenance, PeriodFactEntry,
|
|
219
220
|
create_baseline, get_job_type_economics, list_baselines,
|
|
@@ -293,6 +294,8 @@ __all__ = [
|
|
|
293
294
|
"PeriodFactEntry",
|
|
294
295
|
"Baseline",
|
|
295
296
|
"BaselineProvenance",
|
|
297
|
+
"BaselineThroughputPeriod",
|
|
298
|
+
"BaselineThroughputScope",
|
|
296
299
|
"JobTypeEconomics",
|
|
297
300
|
"JobTypeMetricDirection",
|
|
298
301
|
"JobTypeMonetizationBasis",
|
{revenium_python_sdk-0.14.0 → revenium_python_sdk-0.15.0}/revenium_middleware/_core/cache_tokens.py
RENAMED
|
@@ -13,8 +13,9 @@ fields recurring across multiple emitters) and BACK-1925 for the LiteLLM
|
|
|
13
13
|
client's cache-token contract, which this module's defaults must not break.
|
|
14
14
|
"""
|
|
15
15
|
import logging
|
|
16
|
+
import threading
|
|
16
17
|
from numbers import Number
|
|
17
|
-
from typing import Any, Dict, NamedTuple, Optional
|
|
18
|
+
from typing import Any, Dict, NamedTuple, Optional, Tuple
|
|
18
19
|
|
|
19
20
|
from revenium_middleware._core.package_versions import installed_version, release_of
|
|
20
21
|
|
|
@@ -123,8 +124,8 @@ def billable_input_tokens(
|
|
|
123
124
|
(Anthropic's own ``input_tokens`` spelling reports the cache buckets beside
|
|
124
125
|
it and never reaches this function.)
|
|
125
126
|
|
|
126
|
-
Getting that scope wrong costs money in both directions. Left folded for
|
|
127
|
-
|
|
127
|
+
Getting that scope wrong costs money in both directions. Left folded for a
|
|
128
|
+
response whose cache LiteLLM counted inside the prompt, the cache tokens are priced once at the input rate and
|
|
128
129
|
again in ``cache_read_token_count`` / ``cache_creation_token_count``, which
|
|
129
130
|
is the 4.5x this fix exists for (BACK-3334). Taken out for an OpenAI-shaped
|
|
130
131
|
provider, they are taken out twice: hypercurrent's
|
|
@@ -138,8 +139,10 @@ def billable_input_tokens(
|
|
|
138
139
|
overlap itself, exactly once per record: here for the LiteLLM rows (client
|
|
139
140
|
and gateway), for direct Anthropic and for Claude on Bedrock or Vertex AI,
|
|
140
141
|
and through ``prompt_folds_claude_cache`` for the OpenAI client's records
|
|
141
|
-
of a gateway serving Claude.
|
|
142
|
-
|
|
142
|
+
of a gateway serving Claude. On Bedrock the cache is also removed for every
|
|
143
|
+
call whose usage LiteLLM built with a conversion that counts it inside the
|
|
144
|
+
prompt, whatever the model is called (``prompt_tokens_fold_cache``). Every
|
|
145
|
+
other model's prompt count is reported exactly as LiteLLM built it.
|
|
143
146
|
|
|
144
147
|
A count that comes out negative contradicts the cache counts reported
|
|
145
148
|
beside it, so it is reported unchanged and said out loud rather than
|
|
@@ -163,17 +166,19 @@ ANTHROPIC_MESSAGES_CALL_TYPE = "anthropic_messages"
|
|
|
163
166
|
|
|
164
167
|
# LiteLLM ``custom_llm_provider`` values whose usage conversion folds the cache
|
|
165
168
|
# buckets into ``prompt_tokens`` *and* whose rows the platform stores as sent.
|
|
166
|
-
#
|
|
167
|
-
#
|
|
168
|
-
#
|
|
169
|
-
#
|
|
169
|
+
# Anthropic-shaped usage is the one usage shape Revenium never normalizes
|
|
170
|
+
# server-side (see billable_input_tokens), so it is the only overlap this row
|
|
171
|
+
# may remove: here for direct Anthropic, and in ``_CLAUDE_HOSTING_PROVIDERS``
|
|
172
|
+
# for Claude on Bedrock and Vertex AI, plus the Bedrock conversions that build
|
|
173
|
+
# the same shape whatever the model is named.
|
|
170
174
|
_CACHE_FOLDING_PROVIDERS = frozenset({"anthropic", "anthropic_text"})
|
|
171
175
|
|
|
172
|
-
# Providers that fold the same way when the model they serve is Claude
|
|
173
|
-
# LiteLLM's
|
|
174
|
-
#
|
|
175
|
-
#
|
|
176
|
-
#
|
|
176
|
+
# Providers that fold the same way when the model they serve is Claude, or, on
|
|
177
|
+
# Bedrock, when LiteLLM's own routing picked a conversion that folds: its Vertex
|
|
178
|
+
# AI Claude route and its Bedrock Converse, invoke-Nova and invoke-Anthropic
|
|
179
|
+
# conversions all build ``prompt_tokens`` as uncached input plus cache read plus
|
|
180
|
+
# cache write. Gemini on Vertex AI is reported gross by Google and netted by the
|
|
181
|
+
# platform.
|
|
177
182
|
_CLAUDE_HOSTING_PROVIDERS = frozenset({"bedrock", "vertex_ai"})
|
|
178
183
|
|
|
179
184
|
# Only the top-level fields LiteLLM fills with the deployment it called.
|
|
@@ -250,7 +255,14 @@ def upstream_provider(*sources) -> Optional[str]:
|
|
|
250
255
|
def _provider_folds_cache(provider: str, sources) -> bool:
|
|
251
256
|
if provider in _CACHE_FOLDING_PROVIDERS:
|
|
252
257
|
return True
|
|
253
|
-
|
|
258
|
+
if provider not in _CLAUDE_HOSTING_PROVIDERS:
|
|
259
|
+
return False
|
|
260
|
+
model = _first_named_model(sources)
|
|
261
|
+
if names_claude_model(model):
|
|
262
|
+
return True
|
|
263
|
+
if provider == "bedrock" and model:
|
|
264
|
+
return _bedrock_usage_folds_cache(str(model))
|
|
265
|
+
return False
|
|
254
266
|
|
|
255
267
|
|
|
256
268
|
def prompt_tokens_fold_cache(*sources):
|
|
@@ -281,13 +293,16 @@ def prompt_tokens_fold_cache(*sources):
|
|
|
281
293
|
LiteLLM only ever serves from an Anthropic-shaped upstream, and covers a
|
|
282
294
|
surface carrying the call type but no provider.
|
|
283
295
|
|
|
284
|
-
On ``
|
|
285
|
-
|
|
296
|
+
On ``vertex_ai`` the model decides as well: Claude there is folded like
|
|
297
|
+
Anthropic (BACK-4162), anything else is not. On ``bedrock`` Claude is folded
|
|
298
|
+
by name, and any other model is folded when LiteLLM's own routing for it
|
|
299
|
+
builds usage with a conversion that counts the cache inside the prompt
|
|
300
|
+
(BACK-5780): an application inference profile ARN, Nova. The model is the
|
|
286
301
|
first top-level ``model`` or ``litellm_model_name`` on any source, both
|
|
287
302
|
filled by LiteLLM with the deployment it called. The caller's alias, in the
|
|
288
303
|
request body, its metadata or a gateway's non-streamed ``response.model``,
|
|
289
|
-
is never handed in by a gateway hook, so it cannot turn a
|
|
290
|
-
deployment's count into a
|
|
304
|
+
is never handed in by a gateway hook, so it cannot turn a Gemini
|
|
305
|
+
or Bedrock ``chat_completions`` deployment's count into a netted one.
|
|
291
306
|
|
|
292
307
|
With nothing to read, the prompt count stays as LiteLLM built it: reporting
|
|
293
308
|
a gross count costs the cache overlap twice for Anthropic only, while
|
|
@@ -333,25 +348,84 @@ def litellm_folds_cache_write(converse: bool) -> bool:
|
|
|
333
348
|
return not first <= release < end
|
|
334
349
|
|
|
335
350
|
|
|
336
|
-
|
|
351
|
+
_UNKNOWN = object()
|
|
352
|
+
_INVOKE_PROVIDER_UNKNOWN = object()
|
|
353
|
+
|
|
354
|
+
_BEDROCK_ROUTING_WARNED = False
|
|
355
|
+
_BEDROCK_ROUTING_WARNED_LOCK = threading.Lock()
|
|
356
|
+
|
|
357
|
+
|
|
358
|
+
def _bedrock_usage_family(model: str) -> Tuple[Any, Optional[BaseException]]:
|
|
359
|
+
"""Which usage conversion LiteLLM's own dispatch applies to a Bedrock model.
|
|
360
|
+
|
|
361
|
+
Returns ``(family, error)``. ``family`` is ``"converse"`` or ``"anthropic"``,
|
|
362
|
+
both of which count the cache inside ``prompt_tokens`` and differ in when
|
|
363
|
+
LiteLLM started counting the write; ``None`` for any other route or
|
|
364
|
+
provider; ``_UNKNOWN`` when LiteLLM cannot route at all; or
|
|
365
|
+
``_INVOKE_PROVIDER_UNKNOWN`` for an invoke route whose provider it could not
|
|
366
|
+
name. ``error`` is the exception LiteLLM raised for the last two, else
|
|
367
|
+
``None``. The answer is never cached, so a later call sees LiteLLM as it is
|
|
368
|
+
then.
|
|
369
|
+
"""
|
|
337
370
|
try:
|
|
338
371
|
from litellm.llms.bedrock.common_utils import BedrockModelInfo
|
|
339
372
|
|
|
340
|
-
|
|
341
|
-
except Exception:
|
|
342
|
-
|
|
343
|
-
|
|
373
|
+
route = BedrockModelInfo.get_bedrock_route(model)
|
|
374
|
+
except Exception as error:
|
|
375
|
+
return _UNKNOWN, error
|
|
376
|
+
if route in _BEDROCK_CONVERSE_ROUTES:
|
|
377
|
+
return "converse", None
|
|
378
|
+
if route != "invoke":
|
|
379
|
+
return None, None
|
|
380
|
+
try:
|
|
381
|
+
from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM
|
|
382
|
+
|
|
383
|
+
provider = BaseAWSLLM.get_bedrock_invoke_provider(model)
|
|
384
|
+
except Exception as error:
|
|
385
|
+
return _INVOKE_PROVIDER_UNKNOWN, error
|
|
386
|
+
if provider == "nova":
|
|
387
|
+
return "converse", None
|
|
388
|
+
if provider == "anthropic":
|
|
389
|
+
return "anthropic", None
|
|
390
|
+
return None, None
|
|
391
|
+
|
|
392
|
+
|
|
393
|
+
def _warn_bedrock_routing_unavailable_once(error: BaseException) -> None:
|
|
394
|
+
global _BEDROCK_ROUTING_WARNED
|
|
395
|
+
with _BEDROCK_ROUTING_WARNED_LOCK:
|
|
396
|
+
should_log = not _BEDROCK_ROUTING_WARNED
|
|
397
|
+
_BEDROCK_ROUTING_WARNED = True
|
|
398
|
+
if should_log:
|
|
399
|
+
logger.warning(
|
|
400
|
+
"LiteLLM Bedrock routing unavailable (%s); non-Claude Bedrock rows are "
|
|
401
|
+
"reported with cache tokens inside the input count",
|
|
402
|
+
type(error).__name__,
|
|
403
|
+
)
|
|
404
|
+
|
|
405
|
+
|
|
406
|
+
def _bedrock_usage_folds_cache(model: str) -> bool:
|
|
407
|
+
family, error = _bedrock_usage_family(model)
|
|
408
|
+
if family is _UNKNOWN or family is _INVOKE_PROVIDER_UNKNOWN:
|
|
409
|
+
_warn_bedrock_routing_unavailable_once(error)
|
|
410
|
+
return False
|
|
411
|
+
return family is not None
|
|
344
412
|
|
|
345
413
|
|
|
346
414
|
def routes_via_bedrock_converse(*sources) -> bool:
|
|
347
|
-
"""Whether
|
|
415
|
+
"""Whether the cache write window of this call is Bedrock Converse's.
|
|
348
416
|
|
|
349
417
|
Read from the same sources, in the same order, as ``prompt_tokens_fold_cache``,
|
|
350
|
-
and decided by LiteLLM's own Bedrock routing for the deployment's model
|
|
418
|
+
and decided by LiteLLM's own Bedrock routing for the deployment's model: Nova
|
|
419
|
+
on invoke is built by LiteLLM's Converse conversion, so it counts as Converse.
|
|
420
|
+
When LiteLLM cannot route, Converse is assumed unless the model names invoke.
|
|
351
421
|
"""
|
|
352
422
|
if _upstream_provider(sources) != "bedrock":
|
|
353
423
|
return False
|
|
354
|
-
|
|
424
|
+
model = str(_first_named_model(sources) or "")
|
|
425
|
+
family, _ = _bedrock_usage_family(model)
|
|
426
|
+
if family is _UNKNOWN:
|
|
427
|
+
return "invoke/" not in model
|
|
428
|
+
return family == "converse"
|
|
355
429
|
|
|
356
430
|
|
|
357
431
|
def cache_creation_inside_litellm_prompt(cache_creation_tokens: int, *sources) -> int:
|