revenium-python-sdk 0.11.0__tar.gz → 0.12.1__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-0.11.0 → revenium_python_sdk-0.12.1}/PKG-INFO +17 -6
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/README.md +16 -5
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/pyproject.toml +1 -1
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/enforcement.py +165 -26
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/log_sanitize.py +6 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/metering_buffer.py +1 -5
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/proxy/guardrail.py +57 -13
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/proxy/middleware.py +15 -1
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_python_sdk.egg-info/PKG-INFO +17 -6
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/LICENSE +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/cache_tokens.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/call_context.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/call_ownership.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/closable_scope.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/config.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/context.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/decorators.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/delivery_circuit.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/error_text.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/exceptions.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/failed_calls.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/json_safe.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/langchain_settings.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/load_diagnostics.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/metering.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/metering_pool.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/metering_status.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/metering_submission.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/outcomes.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/package_versions.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/parse_failures.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/patch_registry.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/perplexity_cost.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/perplexity_usage_metadata.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/prompt_extraction.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/responses_status.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/secret_shapes.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/selective_metering.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/shutdown_signals.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/sse_body_tap.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/subscriber.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/tool_event_transport.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/usage_iterations.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/vendor_model_aliases.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/wire_guards.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/LICENSE +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_base_client.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_client.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_compat.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_constants.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_exceptions.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_files.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_models.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_qs.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_resource.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_response.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_streaming.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_types.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_logs.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_proxy.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_reflection.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_resources_proxy.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_streams.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_sync.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_transform.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_typing.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_utils/_utils.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/_version.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/context.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/decorator.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/py.typed +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/resources/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/resources/ai.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/resources/apis.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/resources/events.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/ai_create_audio_params.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/ai_create_completion_params.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/ai_create_image_params.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/ai_create_video_params.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/api_meter_request_params.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/api_meter_response_params.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/event_create_params.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_metering/types/metering_response_resource.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/agentic_outcomes.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/bedrock_adapter.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/bedrock_model_names.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/bedrock_transport.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/config.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/inference_profiles.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/provider.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/raw_response.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/stream_create.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/anthropic/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/fal/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/fal/_call.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/fal/_metering.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/fal/_queue.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/fal/config.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/fal/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/fal/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/common/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/common/exceptions.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/common/protocols.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/common/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/common/types.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/common/utils.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/config.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/google_ai/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/google_ai/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/google_ai/provider.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/vertex_ai/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/vertex_ai/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/google/vertex_ai/provider.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/_metadata.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/_retry.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/anthropic_driver.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/budget_errors.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/litellm_driver.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/ollama_driver.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/openai_driver.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/openai_embedding_driver.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/griptape/universal_driver.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/job_context.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/job_history.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/job_type_economics.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/config.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/context.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/decorators.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/hooks.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/integrations/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/integrations/crewai.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/client/validation.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/proxy/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/litellm/proxy/_metering_owner.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/ollama/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/ollama/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/ollama/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/audio.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/azure_config.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/azure_model_resolver.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/config.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/exceptions.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/_abandoned_runs.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/_model_call.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/_perplexity_cost_capture.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/_prompt_capture.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/_text_completion_usage.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/_utils.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/langchain/unified_handler.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/prompt_extractor.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/provider.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/raw_response.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/openai/xai_usage.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/_metering_scope.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/middleware.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/patching.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/perplexity_sdk.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/provider.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/raw_response.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/streaming.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/perplexity/trace_fields.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/webhooks/__init__.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/webhooks/_verify.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_python_sdk.egg-info/SOURCES.txt +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_python_sdk.egg-info/dependency_links.txt +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_python_sdk.egg-info/requires.txt +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_python_sdk.egg-info/top_level.txt +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/setup.cfg +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/tests/test_metering.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/tests/test_metering_dispatch_timestamps.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/tests/test_patch_coverage.py +0 -0
- {revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/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: 0.12.1
|
|
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
|
|
@@ -192,7 +192,7 @@ Some billable provider calls are not metered at all: see [Calls the SDK does not
|
|
|
192
192
|
The calls below are billed by the provider (some, such as batch and fine-tuning jobs, when the job runs rather than when you call them), but the SDK records no usage for them: nothing reaches Revenium and nothing is logged. On every build a test compares this list with the SDK's inventory of the public calls of the OpenAI, Anthropic, Perplexity, Gemini, Ollama, LiteLLM and fal clients, boto3's Bedrock Runtime client, and the Vertex AI model, session, batch prediction, tuning and context-cache classes, on the provider versions the SDK is tested on. Calls reached another way, for example LangChain classes that call their own vendor SDK, are outside that check, except LangChain embeddings, listed by hand below; wrapping one with `wrap()` logs a warning. Sync and async forms are folded into one name, and the `with_raw_response` and `with_streaming_response` forms of a listed call are not metered either. To record any of these calls, wrap it with `@meter_tool` (see [Tool Metering](#tool-metering)).
|
|
193
193
|
|
|
194
194
|
<!-- not-metered:start -->
|
|
195
|
-
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.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`
|
|
195
|
+
- **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`
|
|
196
196
|
- **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`
|
|
197
197
|
- **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`
|
|
198
198
|
- **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`
|
|
@@ -849,7 +849,7 @@ async def main():
|
|
|
849
849
|
asyncio.run(main())
|
|
850
850
|
```
|
|
851
851
|
|
|
852
|
-
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse` and each
|
|
852
|
+
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse`, each step of `tool_runner` and, on the Anthropic client, each turn of a `thread`), sync and async, including their `with_raw_response` and `with_streaming_response` forms, on the Anthropic, Bedrock and Vertex AI clients. `tool_runner` and `thread` do not accept `usage_metadata=`; set metadata for their calls with `@revenium_metadata` (see [Decorator Support](#decorator-support)). Calls listed under [Calls the SDK does not meter](#calls-the-sdk-does-not-meter), message batches among them, are not metered.
|
|
853
853
|
|
|
854
854
|
A call the provider refuses sends one record with stop reason `ERROR`, zero tokens and the error's class and HTTP status as its error reason (plus the provider's redacted error message when both `REVENIUM_CAPTURE_ERROR_MESSAGES` and `REVENIUM_CAPTURE_PROMPTS` are `true`; a provider's error can quote the request, so it is never sent with prompt capture off; at most 255 characters, keeping its "(HTTP <status>)" ending), as does a `create(stream=True)` stream that breaks off part-way. The HTTP status is also sent as the record's error code; a call that got no HTTP error answer (a connection error, a timeout, or a stream that broke off) has none. Your code still gets the original exception. Calls served through the Bedrock fast path, and calls a spending cap refuses, send no error record.
|
|
855
855
|
|
|
@@ -1209,6 +1209,17 @@ used, because it would make each call a trace of its own, and neither is a sessi
|
|
|
1209
1209
|
id LiteLLM minted itself under its `missing_session_id: generate` policy. The
|
|
1210
1210
|
deprecated callback reads only `x-revenium-trace-id` and `x-claude-code-session-id`.
|
|
1211
1211
|
|
|
1212
|
+
**Trace name.** The guardrail takes a record's trace name from
|
|
1213
|
+
`x-revenium-trace-name`. Without that header it takes the trace name the caller
|
|
1214
|
+
gave LiteLLM, `metadata.trace_name` in the request body (`litellm_metadata.trace_name`
|
|
1215
|
+
on `/v1/messages`), the same field LiteLLM passes to its own logging integrations.
|
|
1216
|
+
Failing both, it uses the gateway's `REVENIUM_TRACE_NAME`, and failing that, the
|
|
1217
|
+
LiteLLM session id that became the trace id (see above), so a session the caller
|
|
1218
|
+
did not title is still named. A call that sends its own `x-revenium-trace-id` is
|
|
1219
|
+
never named after its LiteLLM session. A name over 256 characters is cut to 256. A call
|
|
1220
|
+
with no LiteLLM session and no name has no trace name. The deprecated callback
|
|
1221
|
+
reads only the header and `REVENIUM_TRACE_NAME`.
|
|
1222
|
+
|
|
1212
1223
|
**Organization caps and the organization header.** By default the
|
|
1213
1224
|
`x-revenium-organization-name` header (and the retired `x-revenium-organization-id`)
|
|
1214
1225
|
wins over the calling virtual key's team alias, for the budget check and the record
|
|
@@ -2230,7 +2241,7 @@ pip install 'revenium-python-sdk[openai]'
|
|
|
2230
2241
|
REVENIUM_CIRCUIT_BREAKER_ENABLED=true
|
|
2231
2242
|
REVENIUM_METERING_API_KEY=hak_your_key_here
|
|
2232
2243
|
REVENIUM_TEAM_ID=your_hashed_team_id
|
|
2233
|
-
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
2244
|
+
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional: this is the default
|
|
2234
2245
|
```
|
|
2235
2246
|
|
|
2236
2247
|
### Environment Variables
|
|
@@ -2240,7 +2251,7 @@ REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
|
2240
2251
|
| `REVENIUM_CIRCUIT_BREAKER_ENABLED` | `false` | Master switch. `true` / `1` / `yes` / `on` to enable. |
|
|
2241
2252
|
| `REVENIUM_BYPASS` | `false` | When `true`, every `check_enforcement` call short-circuits to a no-op. Useful for incident response. |
|
|
2242
2253
|
| `REVENIUM_TEAM_ID` | — | Hashed team ID. Path component on rule fetches; required when the breaker is enabled. |
|
|
2243
|
-
| `REVENIUM_ENFORCEMENT_BASE_URL` |
|
|
2254
|
+
| `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. |
|
|
2244
2255
|
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `60` | Background poll interval for rule refreshes. |
|
|
2245
2256
|
| `REVENIUM_CB_FAIL_MODE` | `open` | `open` (default) lets calls through when no cache exists; `closed` raises `BudgetExceededError` until rules are loaded. |
|
|
2246
2257
|
| `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`. |
|
|
@@ -2276,7 +2287,7 @@ When the circuit breaker is disabled, it is a no-op. When enabled:
|
|
|
2276
2287
|
1. A daemon thread (`revenium-enforcement-poll`) starts on first use.
|
|
2277
2288
|
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.
|
|
2278
2289
|
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.
|
|
2279
|
-
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`, however much traffic arrives meanwhile.
|
|
2290
|
+
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`, 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.
|
|
2280
2291
|
5. `204 No Content` is treated as "no rules configured" — the cache is cleared.
|
|
2281
2292
|
|
|
2282
2293
|
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.
|
|
@@ -76,7 +76,7 @@ Some billable provider calls are not metered at all: see [Calls the SDK does not
|
|
|
76
76
|
The calls below are billed by the provider (some, such as batch and fine-tuning jobs, when the job runs rather than when you call them), but the SDK records no usage for them: nothing reaches Revenium and nothing is logged. On every build a test compares this list with the SDK's inventory of the public calls of the OpenAI, Anthropic, Perplexity, Gemini, Ollama, LiteLLM and fal clients, boto3's Bedrock Runtime client, and the Vertex AI model, session, batch prediction, tuning and context-cache classes, on the provider versions the SDK is tested on. Calls reached another way, for example LangChain classes that call their own vendor SDK, are outside that check, except LangChain embeddings, listed by hand below; wrapping one with `wrap()` logs a warning. Sync and async forms are folded into one name, and the `with_raw_response` and `with_streaming_response` forms of a listed call are not metered either. To record any of these calls, wrap it with `@meter_tool` (see [Tool Metering](#tool-metering)).
|
|
77
77
|
|
|
78
78
|
<!-- not-metered:start -->
|
|
79
|
-
- **OpenAI** (`OpenAI` and `AsyncOpenAI` clients): `batches.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`
|
|
79
|
+
- **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`
|
|
80
80
|
- **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`
|
|
81
81
|
- **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`
|
|
82
82
|
- **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`
|
|
@@ -733,7 +733,7 @@ async def main():
|
|
|
733
733
|
asyncio.run(main())
|
|
734
734
|
```
|
|
735
735
|
|
|
736
|
-
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse` and each
|
|
736
|
+
**Note:** The middleware meters `messages.create` (including `create(stream=True)`), `messages.stream` and `messages.parse`, and the same calls on `client.beta.messages` (`create`, `stream`, `parse`, each step of `tool_runner` and, on the Anthropic client, each turn of a `thread`), sync and async, including their `with_raw_response` and `with_streaming_response` forms, on the Anthropic, Bedrock and Vertex AI clients. `tool_runner` and `thread` do not accept `usage_metadata=`; set metadata for their calls with `@revenium_metadata` (see [Decorator Support](#decorator-support)). Calls listed under [Calls the SDK does not meter](#calls-the-sdk-does-not-meter), message batches among them, are not metered.
|
|
737
737
|
|
|
738
738
|
A call the provider refuses sends one record with stop reason `ERROR`, zero tokens and the error's class and HTTP status as its error reason (plus the provider's redacted error message when both `REVENIUM_CAPTURE_ERROR_MESSAGES` and `REVENIUM_CAPTURE_PROMPTS` are `true`; a provider's error can quote the request, so it is never sent with prompt capture off; at most 255 characters, keeping its "(HTTP <status>)" ending), as does a `create(stream=True)` stream that breaks off part-way. The HTTP status is also sent as the record's error code; a call that got no HTTP error answer (a connection error, a timeout, or a stream that broke off) has none. Your code still gets the original exception. Calls served through the Bedrock fast path, and calls a spending cap refuses, send no error record.
|
|
739
739
|
|
|
@@ -1093,6 +1093,17 @@ used, because it would make each call a trace of its own, and neither is a sessi
|
|
|
1093
1093
|
id LiteLLM minted itself under its `missing_session_id: generate` policy. The
|
|
1094
1094
|
deprecated callback reads only `x-revenium-trace-id` and `x-claude-code-session-id`.
|
|
1095
1095
|
|
|
1096
|
+
**Trace name.** The guardrail takes a record's trace name from
|
|
1097
|
+
`x-revenium-trace-name`. Without that header it takes the trace name the caller
|
|
1098
|
+
gave LiteLLM, `metadata.trace_name` in the request body (`litellm_metadata.trace_name`
|
|
1099
|
+
on `/v1/messages`), the same field LiteLLM passes to its own logging integrations.
|
|
1100
|
+
Failing both, it uses the gateway's `REVENIUM_TRACE_NAME`, and failing that, the
|
|
1101
|
+
LiteLLM session id that became the trace id (see above), so a session the caller
|
|
1102
|
+
did not title is still named. A call that sends its own `x-revenium-trace-id` is
|
|
1103
|
+
never named after its LiteLLM session. A name over 256 characters is cut to 256. A call
|
|
1104
|
+
with no LiteLLM session and no name has no trace name. The deprecated callback
|
|
1105
|
+
reads only the header and `REVENIUM_TRACE_NAME`.
|
|
1106
|
+
|
|
1096
1107
|
**Organization caps and the organization header.** By default the
|
|
1097
1108
|
`x-revenium-organization-name` header (and the retired `x-revenium-organization-id`)
|
|
1098
1109
|
wins over the calling virtual key's team alias, for the budget check and the record
|
|
@@ -2114,7 +2125,7 @@ pip install 'revenium-python-sdk[openai]'
|
|
|
2114
2125
|
REVENIUM_CIRCUIT_BREAKER_ENABLED=true
|
|
2115
2126
|
REVENIUM_METERING_API_KEY=hak_your_key_here
|
|
2116
2127
|
REVENIUM_TEAM_ID=your_hashed_team_id
|
|
2117
|
-
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
2128
|
+
REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional: this is the default
|
|
2118
2129
|
```
|
|
2119
2130
|
|
|
2120
2131
|
### Environment Variables
|
|
@@ -2124,7 +2135,7 @@ REVENIUM_ENFORCEMENT_BASE_URL=https://api.revenium.ai/profitstream # optional
|
|
|
2124
2135
|
| `REVENIUM_CIRCUIT_BREAKER_ENABLED` | `false` | Master switch. `true` / `1` / `yes` / `on` to enable. |
|
|
2125
2136
|
| `REVENIUM_BYPASS` | `false` | When `true`, every `check_enforcement` call short-circuits to a no-op. Useful for incident response. |
|
|
2126
2137
|
| `REVENIUM_TEAM_ID` | — | Hashed team ID. Path component on rule fetches; required when the breaker is enabled. |
|
|
2127
|
-
| `REVENIUM_ENFORCEMENT_BASE_URL` |
|
|
2138
|
+
| `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. |
|
|
2128
2139
|
| `REVENIUM_CB_POLL_INTERVAL_SECONDS` | `60` | Background poll interval for rule refreshes. |
|
|
2129
2140
|
| `REVENIUM_CB_FAIL_MODE` | `open` | `open` (default) lets calls through when no cache exists; `closed` raises `BudgetExceededError` until rules are loaded. |
|
|
2130
2141
|
| `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`. |
|
|
@@ -2160,7 +2171,7 @@ When the circuit breaker is disabled, it is a no-op. When enabled:
|
|
|
2160
2171
|
1. A daemon thread (`revenium-enforcement-poll`) starts on first use.
|
|
2161
2172
|
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.
|
|
2162
2173
|
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.
|
|
2163
|
-
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`, however much traffic arrives meanwhile.
|
|
2174
|
+
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`, 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.
|
|
2164
2175
|
5. `204 No Content` is treated as "no rules configured" — the cache is cleared.
|
|
2165
2176
|
|
|
2166
2177
|
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.
|
|
@@ -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 = "0.12.1"
|
|
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"
|
{revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/enforcement.py
RENAMED
|
@@ -69,13 +69,14 @@ import threading
|
|
|
69
69
|
import time
|
|
70
70
|
from email.utils import parsedate_to_datetime
|
|
71
71
|
from typing import Dict, List, NamedTuple, Optional, Set, Tuple
|
|
72
|
-
from urllib.parse import quote,
|
|
72
|
+
from urllib.parse import quote, urlsplit, urlunsplit
|
|
73
73
|
|
|
74
74
|
import httpx
|
|
75
75
|
|
|
76
76
|
from .call_ownership import peek_callback_metadata
|
|
77
77
|
from .closable_scope import ClosableScope
|
|
78
|
-
from .config import Config, env_choice, env_flag_enabled
|
|
78
|
+
from .config import Config, env_choice, env_flag_enabled, get_profitstream_base_url
|
|
79
|
+
from .log_sanitize import address_without_credentials
|
|
79
80
|
from .context import merge_metadata, overlay_metadata
|
|
80
81
|
from .exceptions import BudgetExceededError
|
|
81
82
|
from .fields import extract_common_metadata, extract_organization_name, extract_product_name
|
|
@@ -462,21 +463,73 @@ def _process_metering_url() -> Optional[str]:
|
|
|
462
463
|
return _valid_base_url(str(client.base_url)) if client is not None else None
|
|
463
464
|
|
|
464
465
|
|
|
466
|
+
_PROFITSTREAM_PATH = "/profitstream"
|
|
467
|
+
|
|
468
|
+
# Configured enforcement address -> the sibling address (with or without
|
|
469
|
+
# ``/profitstream``) that served the rules after the configured one answered
|
|
470
|
+
# 404. The cache owner and the snapshot scope keep the configured address;
|
|
471
|
+
# only the request URL moves (BACK-5240).
|
|
472
|
+
_recovered_enforcement_addresses: Dict[str, str] = {}
|
|
473
|
+
|
|
474
|
+
|
|
465
475
|
def _get_enforcement_base_url() -> str:
|
|
466
|
-
"""
|
|
476
|
+
"""The enforcement address as configured: the one the cache belongs to.
|
|
467
477
|
|
|
468
|
-
|
|
469
|
-
(``http://localhost:8080/profitstream``) survives intact.
|
|
470
|
-
the
|
|
471
|
-
|
|
478
|
+
``REVENIUM_ENFORCEMENT_BASE_URL`` is taken as given, so a context path
|
|
479
|
+
(``http://localhost:8080/profitstream``) survives intact. Otherwise it is
|
|
480
|
+
the address outcomes and jobs use (``REVENIUM_PROFITSTREAM_BASE_URL``,
|
|
481
|
+
else the Revenium address the metering client's URL sits under, set by
|
|
482
|
+
``initialize_metering()`` or ``REVENIUM_METERING_BASE_URL``) plus
|
|
483
|
+
``/profitstream``, where the hosted API serves the enforcement rules.
|
|
472
484
|
"""
|
|
473
485
|
explicit = os.environ.get(Config.ENV_REVENIUM_ENFORCEMENT_BASE_URL, "")
|
|
474
486
|
if explicit:
|
|
475
487
|
return explicit.rstrip("/")
|
|
476
|
-
|
|
477
|
-
|
|
478
|
-
|
|
479
|
-
|
|
488
|
+
revenium_address = get_profitstream_base_url(metering_base_url=_process_metering_url())
|
|
489
|
+
return revenium_address.rstrip("/") + _PROFITSTREAM_PATH
|
|
490
|
+
|
|
491
|
+
|
|
492
|
+
def _sibling_enforcement_address(address: str) -> str:
|
|
493
|
+
"""The same address with ``/profitstream`` removed if present, added if not."""
|
|
494
|
+
parts = urlsplit(address)
|
|
495
|
+
path = parts.path.rstrip("/")
|
|
496
|
+
if path.lower().endswith(_PROFITSTREAM_PATH):
|
|
497
|
+
path = path[: -len(_PROFITSTREAM_PATH)]
|
|
498
|
+
else:
|
|
499
|
+
path = path + _PROFITSTREAM_PATH
|
|
500
|
+
return urlunsplit(parts._replace(path=path)).rstrip("/")
|
|
501
|
+
|
|
502
|
+
|
|
503
|
+
def _effective_enforcement_address(configured: str) -> str:
|
|
504
|
+
return _recovered_enforcement_addresses.get(configured, configured)
|
|
505
|
+
|
|
506
|
+
|
|
507
|
+
def _loggable_address(address: str) -> str:
|
|
508
|
+
"""An address as the logs may show it: no credentials, query or fragment."""
|
|
509
|
+
try:
|
|
510
|
+
return address_without_credentials(httpx.URL(address)).rstrip("/")
|
|
511
|
+
except Exception: # noqa: BLE001 - an address httpx cannot parse is not worth a traceback
|
|
512
|
+
return "the configured enforcement address"
|
|
513
|
+
|
|
514
|
+
|
|
515
|
+
def _adopt_enforcement_address(configured: str, tried: str, served: str) -> None:
|
|
516
|
+
"""Send every later read for ``configured`` to ``served``, which just proved it serves the rules."""
|
|
517
|
+
if served == configured:
|
|
518
|
+
_recovered_enforcement_addresses.pop(configured, None)
|
|
519
|
+
else:
|
|
520
|
+
_recovered_enforcement_addresses[configured] = served
|
|
521
|
+
logger.warning(
|
|
522
|
+
"Enforcement rules were not found at %s (HTTP 404) but are served at %s; using that "
|
|
523
|
+
"address from now on. Set %s to it to skip this lookup.",
|
|
524
|
+
_loggable_address(tried), _loggable_address(served), Config.ENV_REVENIUM_ENFORCEMENT_BASE_URL,
|
|
525
|
+
)
|
|
526
|
+
|
|
527
|
+
|
|
528
|
+
def _team_rules_url(address: str, team_id: str) -> str:
|
|
529
|
+
# Percent-encode the team_id path segment so a misconfigured value
|
|
530
|
+
# containing '/', '..', or '?' cannot retarget the request to a
|
|
531
|
+
# different endpoint on the same origin.
|
|
532
|
+
return f"{address}/v2/api/ai/enforcement-rules/{quote(team_id, safe='')}"
|
|
480
533
|
|
|
481
534
|
|
|
482
535
|
class _DepartmentBudgets(NamedTuple):
|
|
@@ -969,6 +1022,9 @@ class _FetchTarget(NamedTuple):
|
|
|
969
1022
|
api_key: str
|
|
970
1023
|
team_url: str
|
|
971
1024
|
owner: Tuple[str, str]
|
|
1025
|
+
# The address ``team_url`` was built on: the configured one, or the
|
|
1026
|
+
# sibling an earlier 404 recovered to.
|
|
1027
|
+
address: str
|
|
972
1028
|
|
|
973
1029
|
|
|
974
1030
|
def _fetch_target() -> Optional[_FetchTarget]:
|
|
@@ -1007,19 +1063,26 @@ def _fetch_target() -> Optional[_FetchTarget]:
|
|
|
1007
1063
|
return None
|
|
1008
1064
|
|
|
1009
1065
|
base_url = _get_enforcement_base_url()
|
|
1010
|
-
|
|
1011
|
-
|
|
1012
|
-
# different endpoint on the same origin.
|
|
1013
|
-
safe_team_id = quote(team_id, safe="")
|
|
1014
|
-
return _FetchTarget(api_key, f"{base_url}/v2/api/ai/enforcement-rules/{safe_team_id}",
|
|
1015
|
-
(base_url, team_id))
|
|
1066
|
+
address = _effective_enforcement_address(base_url)
|
|
1067
|
+
return _FetchTarget(api_key, _team_rules_url(address, team_id), (base_url, team_id), address)
|
|
1016
1068
|
|
|
1017
1069
|
|
|
1018
1070
|
def _get_enforcement(url: str, api_key: str, resource: str,
|
|
1019
1071
|
params: Optional[dict], retry_after_budget: float,
|
|
1020
1072
|
owner: Tuple[str, str]) -> Optional["httpx.Response"]:
|
|
1073
|
+
"""``_get_enforcement_spending`` for a read that has no second address to spend the rest on."""
|
|
1074
|
+
return _get_enforcement_spending(url, api_key, resource, params, retry_after_budget, owner)[0]
|
|
1075
|
+
|
|
1076
|
+
|
|
1077
|
+
def _get_enforcement_spending(url: str, api_key: str, resource: str,
|
|
1078
|
+
params: Optional[dict], retry_after_budget: float,
|
|
1079
|
+
owner: Tuple[str, str]) -> Tuple[Optional["httpx.Response"], float]:
|
|
1021
1080
|
"""GET one enforcement resource under the shared retry and Retry-After budget.
|
|
1022
1081
|
|
|
1082
|
+
Returns the response and the Retry-After budget this read left unspent, so
|
|
1083
|
+
a read that goes on to a second address (``_rules_at_sibling_address``)
|
|
1084
|
+
spends one budget across both and never doubles the documented wait.
|
|
1085
|
+
|
|
1023
1086
|
Returns the first response the retry policy does not retry -- the caller
|
|
1024
1087
|
reads its status and its body -- or ``None`` once the fetch has given up,
|
|
1025
1088
|
so every caller fails open on what it already holds. ``resource`` names
|
|
@@ -1051,10 +1114,10 @@ def _get_enforcement(url: str, api_key: str, resource: str,
|
|
|
1051
1114
|
continue
|
|
1052
1115
|
except Exception:
|
|
1053
1116
|
logger.debug("Failed to fetch enforcement %s, falling open", resource, exc_info=True)
|
|
1054
|
-
return None
|
|
1117
|
+
return None, retry_after_budget
|
|
1055
1118
|
|
|
1056
1119
|
if not _is_retryable_status(response.status_code):
|
|
1057
|
-
return response
|
|
1120
|
+
return response, retry_after_budget
|
|
1058
1121
|
|
|
1059
1122
|
plan = _plan_retry(response, backoff, retry_after_budget, budget_total, owner)
|
|
1060
1123
|
retry_after_budget = plan.budget_remaining
|
|
@@ -1062,7 +1125,7 @@ def _get_enforcement(url: str, api_key: str, resource: str,
|
|
|
1062
1125
|
# Give up without waiting and without retrying: the cooldown
|
|
1063
1126
|
# _plan_retry recorded keeps the server's interval, and the caller
|
|
1064
1127
|
# falls open on the reading it already has.
|
|
1065
|
-
return None
|
|
1128
|
+
return None, retry_after_budget
|
|
1066
1129
|
if attempt == _FETCH_MAX_ATTEMPTS - 1:
|
|
1067
1130
|
break
|
|
1068
1131
|
if _sleep(plan.wait_seconds):
|
|
@@ -1075,7 +1138,7 @@ def _get_enforcement(url: str, api_key: str, resource: str,
|
|
|
1075
1138
|
"Enforcement %s fetch exhausted %d attempts; failing open",
|
|
1076
1139
|
resource, _FETCH_MAX_ATTEMPTS,
|
|
1077
1140
|
)
|
|
1078
|
-
return None
|
|
1141
|
+
return None, retry_after_budget
|
|
1079
1142
|
|
|
1080
1143
|
|
|
1081
1144
|
def _rules_query(rule_id: Optional[str], include_advisory: bool) -> Optional[dict]:
|
|
@@ -1121,15 +1184,25 @@ def _fetch_rules(rule_id: Optional[str] = None,
|
|
|
1121
1184
|
if target is None:
|
|
1122
1185
|
return None
|
|
1123
1186
|
|
|
1124
|
-
response =
|
|
1125
|
-
|
|
1187
|
+
response, budget_left = _get_enforcement_spending(target.team_url, target.api_key, "rule",
|
|
1188
|
+
params, retry_after_budget, target.owner)
|
|
1126
1189
|
if response is None:
|
|
1127
1190
|
return None
|
|
1191
|
+
if response.status_code == 404:
|
|
1192
|
+
recovered = _rules_at_sibling_address(target, params, budget_left)
|
|
1193
|
+
if recovered is not None:
|
|
1194
|
+
return recovered.fetched
|
|
1195
|
+
return _rules_from_response(response)
|
|
1196
|
+
|
|
1128
1197
|
|
|
1198
|
+
def _rules_from_response(response: "httpx.Response") -> Optional[_FetchedRules]:
|
|
1199
|
+
"""The rules a rules read answered with, or ``None`` when it answered with anything else."""
|
|
1129
1200
|
try:
|
|
1130
1201
|
# 204 No Content == no rules configured for this team; cache empty
|
|
1131
1202
|
if response.status_code == 204:
|
|
1132
1203
|
return _FetchedRules([], {}, {}, {}, {})
|
|
1204
|
+
if 400 <= response.status_code < 500:
|
|
1205
|
+
_warn_refused_rules_read(response)
|
|
1133
1206
|
response.raise_for_status()
|
|
1134
1207
|
# Server currently returns ``{"rules": [...], "compiledAt": ...,
|
|
1135
1208
|
# "departmentBudgetBlocks": {...}, "departmentBudgetBlockBalances":
|
|
@@ -1141,6 +1214,60 @@ def _fetch_rules(rule_id: Optional[str] = None,
|
|
|
1141
1214
|
return None
|
|
1142
1215
|
|
|
1143
1216
|
|
|
1217
|
+
class _RecoveredRules(NamedTuple):
|
|
1218
|
+
address: str
|
|
1219
|
+
fetched: _FetchedRules
|
|
1220
|
+
budget_left: float
|
|
1221
|
+
|
|
1222
|
+
|
|
1223
|
+
def _rules_at_sibling_address(target: _FetchTarget, params: Optional[dict],
|
|
1224
|
+
retry_after_budget: float) -> Optional[_RecoveredRules]:
|
|
1225
|
+
"""Read the rules once at the sibling of the address that answered 404, and adopt it if it serves them.
|
|
1226
|
+
|
|
1227
|
+
The hosted API serves enforcement under ``/profitstream``; a self-hosted
|
|
1228
|
+
server may serve it at the root, and an operator may have configured
|
|
1229
|
+
either address without that segment. A 404 says the rules are not at this
|
|
1230
|
+
address at all, not that the team has none (that is a 204), so the one
|
|
1231
|
+
other address is tried. It is adopted for every later read only once it
|
|
1232
|
+
has answered with rules (a 204, or a body ``_fetched_from_payload``
|
|
1233
|
+
accepts): a refusal or an unreadable body there leaves the configured
|
|
1234
|
+
address in use, so the next poll reads it again. A refusal at the sibling
|
|
1235
|
+
(any 4xx but 404) is logged by ``_warn_refused_rules_read`` as well, so a
|
|
1236
|
+
wrong key or team id at the working address is not hidden behind the
|
|
1237
|
+
first address's 404.
|
|
1238
|
+
"""
|
|
1239
|
+
configured, team_id = target.owner
|
|
1240
|
+
sibling = _sibling_enforcement_address(target.address)
|
|
1241
|
+
response, budget_left = _get_enforcement_spending(
|
|
1242
|
+
_team_rules_url(sibling, team_id), target.api_key, "rule", params,
|
|
1243
|
+
retry_after_budget, target.owner)
|
|
1244
|
+
if response is None:
|
|
1245
|
+
return None
|
|
1246
|
+
if not (response.is_success or response.status_code == 204):
|
|
1247
|
+
if 400 <= response.status_code < 500 and response.status_code != 404:
|
|
1248
|
+
_warn_refused_rules_read(response)
|
|
1249
|
+
return None
|
|
1250
|
+
fetched = _rules_from_response(response)
|
|
1251
|
+
if fetched is None:
|
|
1252
|
+
return None
|
|
1253
|
+
_adopt_enforcement_address(configured, target.address, sibling)
|
|
1254
|
+
return _RecoveredRules(sibling, fetched, budget_left)
|
|
1255
|
+
|
|
1256
|
+
|
|
1257
|
+
def _warn_refused_rules_read(response: "httpx.Response") -> None:
|
|
1258
|
+
"""Say which URL refused the rules read and why, so a wrong address or key is visible."""
|
|
1259
|
+
status = response.status_code
|
|
1260
|
+
if status == 404:
|
|
1261
|
+
hint = (" The enforcement API is not at this address: set %s to the Revenium address "
|
|
1262
|
+
"that serves /v2/api/ai/enforcement-rules." % Config.ENV_REVENIUM_ENFORCEMENT_BASE_URL)
|
|
1263
|
+
elif status in (401, 403):
|
|
1264
|
+
hint = " Check %s and %s." % (Config.ENV_REVENIUM_API_KEY, Config.ENV_REVENIUM_TEAM_ID)
|
|
1265
|
+
else:
|
|
1266
|
+
hint = ""
|
|
1267
|
+
logger.warning("Enforcement rules could not be read: GET %s answered HTTP %d.%s",
|
|
1268
|
+
address_without_credentials(response.request.url), status, hint)
|
|
1269
|
+
|
|
1270
|
+
|
|
1144
1271
|
def _fetch_roster(rule_id: str, page: int = 0, size: int = 25,
|
|
1145
1272
|
search: Optional[str] = None,
|
|
1146
1273
|
band: Optional[str] = None) -> Optional[dict]:
|
|
@@ -1172,11 +1299,23 @@ def _fetch_roster(rule_id: str, page: int = 0, size: int = 25,
|
|
|
1172
1299
|
if band:
|
|
1173
1300
|
params["band"] = band
|
|
1174
1301
|
|
|
1175
|
-
response =
|
|
1176
|
-
|
|
1177
|
-
|
|
1302
|
+
response, budget_left = _get_enforcement_spending(
|
|
1303
|
+
f"{target.team_url}/roster", target.api_key, "roster", params,
|
|
1304
|
+
_INSPECTION_RETRY_AFTER_GIVE_UP_SECONDS, target.owner)
|
|
1178
1305
|
if response is None:
|
|
1179
1306
|
return None
|
|
1307
|
+
if response.status_code == 404:
|
|
1308
|
+
# Either the rule has no roster or the rules are not at this address.
|
|
1309
|
+
# Only a rules read can tell: if it recovers the address, the roster
|
|
1310
|
+
# is read once more there.
|
|
1311
|
+
recovered = _rules_at_sibling_address(target, None, budget_left)
|
|
1312
|
+
if recovered is None:
|
|
1313
|
+
return None
|
|
1314
|
+
response = _get_enforcement(f"{_team_rules_url(recovered.address, target.owner[1])}/roster",
|
|
1315
|
+
target.api_key, "roster", params, recovered.budget_left,
|
|
1316
|
+
target.owner)
|
|
1317
|
+
if response is None:
|
|
1318
|
+
return None
|
|
1180
1319
|
|
|
1181
1320
|
try:
|
|
1182
1321
|
# 204 No Content == the rule has no compiled reading yet, which is a
|
{revenium_python_sdk-0.11.0 → revenium_python_sdk-0.12.1}/revenium_middleware/_core/log_sanitize.py
RENAMED
|
@@ -264,3 +264,9 @@ def redact_error_text(text: str, max_len: int) -> str:
|
|
|
264
264
|
redacted = _EMAIL_WORD_PATTERN.sub(_REDACTED, redacted)
|
|
265
265
|
redacted = _SECRET_SHAPED_WORD_PATTERN.sub(_REDACTED, redacted)
|
|
266
266
|
return redacted[:max_len]
|
|
267
|
+
|
|
268
|
+
|
|
269
|
+
def address_without_credentials(url) -> str:
|
|
270
|
+
"""Scheme, host, port and path of an ``httpx.URL``: never its userinfo, query or fragment."""
|
|
271
|
+
port = f":{url.port}" if url.port is not None else ""
|
|
272
|
+
return f"{url.scheme}://{url.host}{port}{url.path}"
|
|
@@ -29,6 +29,7 @@ from typing import Any, Callable, Deque, Dict, Iterator, List, NamedTuple, Optio
|
|
|
29
29
|
import httpx
|
|
30
30
|
|
|
31
31
|
from revenium_middleware._core.config import read_env_number
|
|
32
|
+
from revenium_middleware._core.log_sanitize import address_without_credentials as _address_without_credentials
|
|
32
33
|
from revenium_middleware._core.delivery_circuit import get_circuit
|
|
33
34
|
from revenium_middleware._core.metering_status import (
|
|
34
35
|
record_metering_error,
|
|
@@ -378,11 +379,6 @@ def is_retryable_failure(exc: BaseException) -> bool:
|
|
|
378
379
|
return False
|
|
379
380
|
|
|
380
381
|
|
|
381
|
-
def _address_without_credentials(url: httpx.URL) -> str:
|
|
382
|
-
port = f":{url.port}" if url.port is not None else ""
|
|
383
|
-
return f"{url.scheme}://{url.host}{port}{url.path}"
|
|
384
|
-
|
|
385
|
-
|
|
386
382
|
def _redirect_target(response: httpx.Response) -> str:
|
|
387
383
|
location = response.headers.get("location")
|
|
388
384
|
if not location:
|
|
@@ -786,6 +786,26 @@ def _litellm_session_id(container):
|
|
|
786
786
|
return None
|
|
787
787
|
|
|
788
788
|
|
|
789
|
+
def _litellm_trace_name(container):
|
|
790
|
+
"""The trace name the caller gave LiteLLM, or None.
|
|
791
|
+
|
|
792
|
+
LiteLLM takes a trace name as ``metadata.trace_name``, one of its trace
|
|
793
|
+
control fields alongside ``session_id``, and on the routes that use
|
|
794
|
+
``litellm_metadata`` it promotes the caller's value there. ``container`` is
|
|
795
|
+
the same per-hook request surface ``_litellm_session_id`` reads, and
|
|
796
|
+
``litellm_metadata`` is read first for the same reason.
|
|
797
|
+
|
|
798
|
+
Only a non-blank string counts, because this runs in-band and a non-string
|
|
799
|
+
must never reach the wire.
|
|
800
|
+
"""
|
|
801
|
+
container = _as_mapping(container)
|
|
802
|
+
for key in ("litellm_metadata", "metadata"):
|
|
803
|
+
candidate = _as_mapping(container.get(key)).get("trace_name")
|
|
804
|
+
if isinstance(candidate, str) and present_or_none(candidate) is not None:
|
|
805
|
+
return candidate
|
|
806
|
+
return None
|
|
807
|
+
|
|
808
|
+
|
|
789
809
|
def _organization_name(data, headers, metadata, user_api_key_dict, key_metadata):
|
|
790
810
|
"""Organization attribution, headers first then key/team identity.
|
|
791
811
|
|
|
@@ -1399,6 +1419,7 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1399
1419
|
mediation_latency=mediation_latency,
|
|
1400
1420
|
stop_reason="END",
|
|
1401
1421
|
litellm_session_id=_litellm_session_id(data),
|
|
1422
|
+
litellm_trace_name=_litellm_trace_name(data),
|
|
1402
1423
|
# LiteLLM-populated surfaces only: data["metadata"] (and so
|
|
1403
1424
|
# hidden_params) is the caller's own request body on
|
|
1404
1425
|
# /v1/messages, and a caller must not be able to name the
|
|
@@ -1545,6 +1566,7 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1545
1566
|
# litellm_params, not kwargs: LiteLLM spreads provider params
|
|
1546
1567
|
# across the kwargs' top level, which the caller can shape.
|
|
1547
1568
|
litellm_session_id=_litellm_session_id(litellm_params),
|
|
1569
|
+
litellm_trace_name=_litellm_trace_name(litellm_params),
|
|
1548
1570
|
upstream_sources=(
|
|
1549
1571
|
getattr(response_obj, "_hidden_params", None),
|
|
1550
1572
|
kwargs,
|
|
@@ -1675,6 +1697,7 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1675
1697
|
mediation_latency=hidden_params.get("litellm_overhead_time_ms") or 0,
|
|
1676
1698
|
stop_reason="ERROR",
|
|
1677
1699
|
litellm_session_id=_litellm_session_id(request_data),
|
|
1700
|
+
litellm_trace_name=_litellm_trace_name(request_data),
|
|
1678
1701
|
upstream_sources=(
|
|
1679
1702
|
getattr(original_exception, "_hidden_params", None),
|
|
1680
1703
|
request_data.get("litellm_params"),
|
|
@@ -1712,6 +1735,7 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1712
1735
|
upstream_sources,
|
|
1713
1736
|
is_streamed=None,
|
|
1714
1737
|
litellm_session_id=None,
|
|
1738
|
+
litellm_trace_name=None,
|
|
1715
1739
|
):
|
|
1716
1740
|
"""Build the completion metering payload shared by all three hooks.
|
|
1717
1741
|
|
|
@@ -1740,6 +1764,12 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1740
1764
|
``x-revenium-trace-id`` header was sent. Resolved by the hook
|
|
1741
1765
|
because the logging path's ``data`` is not a LiteLLM-owned
|
|
1742
1766
|
surface at its top level.
|
|
1767
|
+
litellm_trace_name: The trace name the caller gave LiteLLM,
|
|
1768
|
+
resolved by the hook with ``_litellm_trace_name``. It is the
|
|
1769
|
+
trace name when no ``x-revenium-trace-name`` header was sent,
|
|
1770
|
+
and outranks the host's ``REVENIUM_TRACE_NAME``. When none of
|
|
1771
|
+
the three is set, ``litellm_session_id`` names the trace, but
|
|
1772
|
+
only when it is also the trace id.
|
|
1743
1773
|
|
|
1744
1774
|
Returns:
|
|
1745
1775
|
A ``(completion_args, extra_body)`` tuple -- ``extra_body`` carries
|
|
@@ -1766,6 +1796,31 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1766
1796
|
# is what keeps the published "Z" honest for both.
|
|
1767
1797
|
response_time_str = format_utc_timestamp(response_time)
|
|
1768
1798
|
|
|
1799
|
+
# Order: the x-revenium-trace-id header, the LiteLLM session id the
|
|
1800
|
+
# caller supplied, then Claude Code's session id. A session id groups
|
|
1801
|
+
# calls into one trace, and Claude Code stamps its own on its telemetry
|
|
1802
|
+
# rows as their trace id, so one session reads as one trace on both
|
|
1803
|
+
# records. LiteLLM's own per-request trace id is never a fallback: it
|
|
1804
|
+
# mints a fresh uuid4 for every request, which would make every call
|
|
1805
|
+
# its own trace. Never the transaction id: a session covers many calls,
|
|
1806
|
+
# and keying the duplicate gate on it would fold a whole session into
|
|
1807
|
+
# one record and under-bill it.
|
|
1808
|
+
trace_id = (headers.get("x-revenium-trace-id")
|
|
1809
|
+
or litellm_session_id
|
|
1810
|
+
or headers.get("x-claude-code-session-id"))
|
|
1811
|
+
# The session id names the trace only when it is the trace: under a
|
|
1812
|
+
# caller's own x-revenium-trace-id it would label that trace with an id
|
|
1813
|
+
# it does not carry.
|
|
1814
|
+
trace_fields = trace_fields_from_headers(
|
|
1815
|
+
headers,
|
|
1816
|
+
litellm_trace_name=litellm_trace_name,
|
|
1817
|
+
default_trace_name=(
|
|
1818
|
+
litellm_session_id
|
|
1819
|
+
if litellm_session_id and trace_id == litellm_session_id
|
|
1820
|
+
else None
|
|
1821
|
+
),
|
|
1822
|
+
)
|
|
1823
|
+
|
|
1769
1824
|
completion_args = {
|
|
1770
1825
|
"cache_creation_token_count": cache_creation_tokens,
|
|
1771
1826
|
**extract_cache_creation_ttl_counts(usage),
|
|
@@ -1788,18 +1843,7 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1788
1843
|
"stop_reason": stop_reason,
|
|
1789
1844
|
"total_token_count": total_tokens,
|
|
1790
1845
|
"transaction_id": transaction_id,
|
|
1791
|
-
|
|
1792
|
-
# caller supplied, then Claude Code's session id. A session id
|
|
1793
|
-
# groups calls into one trace, and Claude Code stamps its own on
|
|
1794
|
-
# its telemetry rows as their trace id, so one session reads as one
|
|
1795
|
-
# trace on both records. LiteLLM's own per-request trace id is
|
|
1796
|
-
# never a fallback: it mints a fresh uuid4 for every request, which
|
|
1797
|
-
# would make every call its own trace. Never the transaction id: a
|
|
1798
|
-
# session covers many calls, and keying the duplicate gate on it
|
|
1799
|
-
# would fold a whole session into one record and under-bill it.
|
|
1800
|
-
"trace_id": (headers.get("x-revenium-trace-id")
|
|
1801
|
-
or litellm_session_id
|
|
1802
|
-
or headers.get("x-claude-code-session-id")),
|
|
1846
|
+
"trace_id": trace_id,
|
|
1803
1847
|
"task_type": headers.get("x-revenium-task-type"),
|
|
1804
1848
|
"subscriber": subscriber if subscriber else None,
|
|
1805
1849
|
"organization_name": _organization_name(
|
|
@@ -1820,7 +1864,7 @@ class ReveniumGuardrail(CustomGuardrail):
|
|
|
1820
1864
|
"is_streamed": bool(is_streamed),
|
|
1821
1865
|
"operation_type": operation_type,
|
|
1822
1866
|
"mediation_latency": mediation_latency,
|
|
1823
|
-
**
|
|
1867
|
+
**trace_fields,
|
|
1824
1868
|
"middleware_source": "GUARDRAIL",
|
|
1825
1869
|
}
|
|
1826
1870
|
extra_body = merge_extra_body(None, _job_fields(headers, key_metadata))
|