chore(release): backport #33810, #33733 to rc/1.94.0 and bump litellm-proxy-extras to 0.4.79.post1

litellm-proxy-extras/litellm_proxy_extras/migrations/20260717000000_add_compression_saved_tokens/migration.sql

@@ -0,0 +1,17 @@
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyUserSpend" ADD COLUMN IF NOT EXISTS "compression_saved_tokens" BIGINT NOT NULL DEFAULT 0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyOrganizationSpend" ADD COLUMN IF NOT EXISTS "compression_saved_tokens" BIGINT NOT NULL DEFAULT 0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyEndUserSpend" ADD COLUMN IF NOT EXISTS "compression_saved_tokens" BIGINT NOT NULL DEFAULT 0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyAgentSpend" ADD COLUMN IF NOT EXISTS "compression_saved_tokens" BIGINT NOT NULL DEFAULT 0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyTeamSpend" ADD COLUMN IF NOT EXISTS "compression_saved_tokens" BIGINT NOT NULL DEFAULT 0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyTagSpend" ADD COLUMN IF NOT EXISTS "compression_saved_tokens" BIGINT NOT NULL DEFAULT 0;

litellm-proxy-extras/litellm_proxy_extras/migrations/20260718000000_add_savings_spend/migration.sql

@@ -0,0 +1,23 @@
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyUserSpend" ADD COLUMN IF NOT EXISTS "compression_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+ALTER TABLE "LiteLLM_DailyUserSpend" ADD COLUMN IF NOT EXISTS "prompt_caching_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyOrganizationSpend" ADD COLUMN IF NOT EXISTS "compression_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+ALTER TABLE "LiteLLM_DailyOrganizationSpend" ADD COLUMN IF NOT EXISTS "prompt_caching_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyEndUserSpend" ADD COLUMN IF NOT EXISTS "compression_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+ALTER TABLE "LiteLLM_DailyEndUserSpend" ADD COLUMN IF NOT EXISTS "prompt_caching_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyAgentSpend" ADD COLUMN IF NOT EXISTS "compression_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+ALTER TABLE "LiteLLM_DailyAgentSpend" ADD COLUMN IF NOT EXISTS "prompt_caching_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyTeamSpend" ADD COLUMN IF NOT EXISTS "compression_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+ALTER TABLE "LiteLLM_DailyTeamSpend" ADD COLUMN IF NOT EXISTS "prompt_caching_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+
+-- AlterTable
+ALTER TABLE "LiteLLM_DailyTagSpend" ADD COLUMN IF NOT EXISTS "compression_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;
+ALTER TABLE "LiteLLM_DailyTagSpend" ADD COLUMN IF NOT EXISTS "prompt_caching_savings_spend" DOUBLE PRECISION NOT NULL DEFAULT 0.0;

litellm-proxy-extras/litellm_proxy_extras/schema.prisma

@@ -736,6 +736,9 @@ model LiteLLM_DailyUserSpend {
   completion_tokens   BigInt      @default(0)
   cache_read_input_tokens     BigInt      @default(0)
   cache_creation_input_tokens BigInt      @default(0)
+  compression_saved_tokens    BigInt      @default(0)
+  compression_savings_spend    Float       @default(0.0)
+  prompt_caching_savings_spend Float       @default(0.0)
   spend               Float    @default(0.0)
   api_requests        BigInt      @default(0)
   successful_requests BigInt      @default(0)
@@ -767,6 +770,9 @@ model LiteLLM_DailyOrganizationSpend {
   completion_tokens   BigInt      @default(0)
   cache_read_input_tokens     BigInt      @default(0)
   cache_creation_input_tokens BigInt      @default(0)
+  compression_saved_tokens    BigInt      @default(0)
+  compression_savings_spend    Float       @default(0.0)
+  prompt_caching_savings_spend Float       @default(0.0)
   spend               Float    @default(0.0)
   api_requests        BigInt      @default(0)
   successful_requests BigInt      @default(0)
@@ -798,6 +804,9 @@ model LiteLLM_DailyEndUserSpend {
   completion_tokens   BigInt      @default(0)
   cache_read_input_tokens     BigInt      @default(0)
   cache_creation_input_tokens BigInt      @default(0)
+  compression_saved_tokens    BigInt      @default(0)
+  compression_savings_spend    Float       @default(0.0)
+  prompt_caching_savings_spend Float       @default(0.0)
   spend               Float    @default(0.0)
   api_requests        BigInt      @default(0)
   successful_requests BigInt      @default(0)
@@ -828,6 +837,9 @@ model LiteLLM_DailyAgentSpend {
   completion_tokens   BigInt      @default(0)
   cache_read_input_tokens     BigInt      @default(0)
   cache_creation_input_tokens BigInt      @default(0)
+  compression_saved_tokens    BigInt      @default(0)
+  compression_savings_spend    Float       @default(0.0)
+  prompt_caching_savings_spend Float       @default(0.0)
   spend               Float    @default(0.0)
   api_requests        BigInt      @default(0)
   successful_requests BigInt      @default(0)
@@ -858,6 +870,9 @@ model LiteLLM_DailyTeamSpend {
   completion_tokens   BigInt      @default(0)
   cache_read_input_tokens     BigInt      @default(0)
   cache_creation_input_tokens BigInt      @default(0)
+  compression_saved_tokens    BigInt      @default(0)
+  compression_savings_spend    Float       @default(0.0)
+  prompt_caching_savings_spend Float       @default(0.0)
   spend               Float    @default(0.0)
   api_requests        BigInt      @default(0)
   successful_requests BigInt      @default(0)
@@ -890,6 +905,9 @@ model LiteLLM_DailyTagSpend {
   completion_tokens   BigInt      @default(0)
   cache_read_input_tokens     BigInt      @default(0)
   cache_creation_input_tokens BigInt      @default(0)
+  compression_saved_tokens    BigInt      @default(0)
+  compression_savings_spend    Float       @default(0.0)
+  prompt_caching_savings_spend Float       @default(0.0)
   spend               Float    @default(0.0)
   api_requests        BigInt      @default(0)
   successful_requests BigInt      @default(0)

litellm-proxy-extras/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "litellm-proxy-extras"
-version = "0.4.79"
+version = "0.4.79.post1"
 description = "Additional files for the LiteLLM Proxy. Reduces the size of the main litellm package."
 readme = "README.md"
 requires-python = ">=3.9"
@@ -26,7 +26,7 @@ required-version = ">=0.10.9"
 module-root = ""
 
 [tool.commitizen]
-version = "0.4.79"
+version = "0.4.79.post1"
 version_files = [
     "pyproject.toml:^version",
     "../pyproject.toml:litellm-proxy-extras==",

litellm/integrations/compression_interception/handler.py

@@ -14,6 +14,7 @@
 from litellm.integrations.custom_logger import CustomLogger
 from litellm.types.integrations.compression_interception import (
     CompressionInterceptionConfig,
+    CompressionSavingsMetadata,
 )
 from litellm.types.integrations.custom_logger import (
     AgenticLoopPlan,
@@ -25,6 +26,41 @@
 _CACHE_TTL_SECONDS = 15 * 60
 
 
+def _compression_savings_from_counts(
+    original_tokens: object, compressed_tokens: object
+) -> CompressionSavingsMetadata | None:
+    if isinstance(original_tokens, bool) or not isinstance(original_tokens, int):
+        return None
+    if isinstance(compressed_tokens, bool) or not isinstance(compressed_tokens, int):
+        return None
+    if compressed_tokens < 0 or original_tokens < compressed_tokens:
+        return None
+    return CompressionSavingsMetadata(
+        tokens_before=original_tokens,
+        tokens_after=compressed_tokens,
+        tokens_saved=original_tokens - compressed_tokens,
+        source="compression_interception",
+    )
+
+
+def _record_compression_savings(kwargs: dict[str, object], savings: CompressionSavingsMetadata) -> None:
+    """
+    Attach savings to the request's litellm metadata so they land in the
+    SpendLog row's metadata JSON under ``compression_savings``.
+
+    ``/v1/messages`` requests carry proxy metadata under ``litellm_metadata``
+    (the ``metadata`` key is Anthropic's own API field). The existing dict is
+    updated in place because the proxy and the logging object hold references
+    to the same object; replacing it would orphan writes made through those
+    references.
+    """
+    existing = kwargs.get("litellm_metadata")
+    if isinstance(existing, dict):
+        existing["compression_savings"] = savings
+        return
+    kwargs["litellm_metadata"] = {"compression_savings": savings}
+
+
 class CompressionInterceptionLogger(CustomLogger):
     """
     CustomLogger that implements transparent prompt compression + retrieval loops.
@@ -130,6 +166,12 @@ async def async_pre_call_deployment_hook(
                 call_id = str(uuid.uuid4())
                 kwargs["litellm_call_id"] = call_id
             self._compression_cache_by_call_id[call_id] = (cache, time.time())
+            savings = _compression_savings_from_counts(
+                original_tokens=compressed.get("original_tokens"),
+                compressed_tokens=compressed.get("compressed_tokens"),
+            )
+            if savings is not None:
+                _record_compression_savings(kwargs=kwargs, savings=savings)
             verbose_logger.debug(
                 "CompressionInterception: compressed request [call_id=%s original=%d compressed=%d cached_keys=%d]",
                 call_id,

litellm/llms/bedrock/batches/transformation.py

@@ -4,6 +4,7 @@
 from typing import Any, Dict, List, Literal, Optional, Union, cast
 
 from httpx import Headers, Response
+from pydantic import TypeAdapter, ValidationError
 
 from litellm.litellm_core_utils.cloud_storage_security import (
     BEDROCK_MANAGED_S3_BATCH_PREFIX,
@@ -19,6 +20,7 @@
     BedrockOutputDataConfig,
     BedrockS3InputDataConfig,
     BedrockS3OutputDataConfig,
+    BedrockTag,
 )
 from litellm.types.llms.openai import (
     AllMessageValues,
@@ -38,6 +40,18 @@
     r"-[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}\.jsonl$"
 )
 
+_BEDROCK_TAGS_ADAPTER: TypeAdapter[list[BedrockTag]] = TypeAdapter(list[BedrockTag])
+
+
+def _validate_bedrock_tags(raw_tags: object) -> list[BedrockTag]:
+    try:
+        return _BEDROCK_TAGS_ADAPTER.validate_python(raw_tags, strict=True)
+    except ValidationError as e:
+        raise ValueError(
+            "Invalid 'bedrock_tags' value. Expected a list of {'key': <str>, 'value': <str>} dicts, "
+            f"e.g. [{{'key': 'team', 'value': 'genai'}}]. Got: {raw_tags!r}"
+        ) from e
+
 
 class BedrockBatchesConfig(BaseAWSLLM, BaseBatchesConfig):
     """
@@ -201,6 +215,11 @@ def transform_create_batch_request(
             "roleArn": role_arn,
         }
 
+        config_bedrock_tags = litellm_params.get("bedrock_tags")
+        bedrock_tags = config_bedrock_tags if config_bedrock_tags is not None else optional_params.get("bedrock_tags")
+        if bedrock_tags is not None:
+            bedrock_request["tags"] = _validate_bedrock_tags(bedrock_tags)
+
         # Add optional parameters if provided
         completion_window = create_batch_data.get("completion_window")
         if completion_window:

litellm/proxy/_types.py

@@ -20,6 +20,9 @@
 from litellm.litellm_core_utils.initialize_dynamic_callback_params import (
     validate_no_callback_env_reference,
 )
+from litellm.types.integrations.compression_interception import (
+    CompressionSavingsMetadata,
+)
 from litellm.types.integrations.slack_alerting import AlertType
 from litellm.types.llms.openai import (
     AllMessageValues,
@@ -3242,6 +3245,7 @@ class SpendLogsMetadata(TypedDict):
     attempted_retries: Optional[int]  # Number of retries attempted (0 = first attempt succeeded)
     max_retries: Optional[int]  # Max retries configured for this request
     cost_breakdown: Optional[CostBreakdown]  # Detailed cost breakdown (input_cost, output_cost, margin, discount, etc.)
+    compression_savings: CompressionSavingsMetadata | None
 
 
 class SpendLogsPayload(TypedDict):
@@ -4461,6 +4465,11 @@ class BaseDailySpendTransaction(TypedDict):
     completion_tokens: int
     cache_read_input_tokens: int
     cache_creation_input_tokens: int
+    compression_saved_tokens: int
+
+    # cost-savings metrics (dollars, priced per request before aggregation)
+    compression_savings_spend: float
+    prompt_caching_savings_spend: float
 
     # request level metrics
     spend: float

litellm/proxy/agent_endpoints/endpoints.py

@@ -1082,6 +1082,7 @@ async def get_agent_daily_activity(
                     total_failed_requests=0,
                     total_cache_read_input_tokens=0,
                     total_cache_creation_input_tokens=0,
+                    total_compression_saved_tokens=0,
                     page=page,
                     total_pages=0,
                     has_more=False,

litellm/proxy/auth/auth_utils.py

@@ -273,6 +273,7 @@ def _build_banned_observability_params() -> FrozenSet[str]:
     # re-route the request's retention and accounting to any project
     # reachable with the deployment's shared AWS credentials.
     "aws_bedrock_project_id",
+    "bedrock_tags",
     # Provider-specific endpoint overrides that flow into the outbound
     # request via ``optional_params``. Same threat as ``api_base``:
     # ``s3_endpoint_url`` redirects Bedrock file uploads to attacker

litellm/proxy/db/db_spend_update_writer.py

@@ -59,6 +59,10 @@
     ToolDiscoveryQueue,
 )
 from litellm.proxy.route_llm_request import ROUTE_ENDPOINT_MAPPING
+from litellm.proxy.spend_tracking.compression_savings import (
+    extract_compression_saved_tokens,
+)
+from litellm.proxy.spend_tracking.savings import compute_savings_spend
 from litellm.proxy.spend_tracking.spend_log_error_logger import spend_log_error
 
 if TYPE_CHECKING:
@@ -1554,6 +1558,18 @@ async def _update_daily_spend(
                                         common_data["cache_creation_input_tokens"] = transaction.get(
                                             "cache_creation_input_tokens", 0
                                         )
+                                    if "compression_saved_tokens" in transaction:
+                                        common_data["compression_saved_tokens"] = transaction.get(
+                                            "compression_saved_tokens", 0
+                                        )
+                                    if "compression_savings_spend" in transaction:
+                                        common_data["compression_savings_spend"] = transaction.get(
+                                            "compression_savings_spend", 0
+                                        )
+                                    if "prompt_caching_savings_spend" in transaction:
+                                        common_data["prompt_caching_savings_spend"] = transaction.get(
+                                            "prompt_caching_savings_spend", 0
+                                        )
 
                                     if entity_type == "tag" and "request_id" in transaction:
                                         common_data["request_id"] = transaction.get("request_id")
@@ -1577,6 +1593,18 @@ async def _update_daily_spend(
                                         update_data["cache_creation_input_tokens"] = {
                                             "increment": transaction.get("cache_creation_input_tokens", 0)
                                         }
+                                    if "compression_saved_tokens" in transaction:
+                                        update_data["compression_saved_tokens"] = {
+                                            "increment": transaction.get("compression_saved_tokens", 0)
+                                        }
+                                    if "compression_savings_spend" in transaction:
+                                        update_data["compression_savings_spend"] = {
+                                            "increment": transaction.get("compression_savings_spend", 0)
+                                        }
+                                    if "prompt_caching_savings_spend" in transaction:
+                                        update_data["prompt_caching_savings_spend"] = {
+                                            "increment": transaction.get("prompt_caching_savings_spend", 0)
+                                        }
 
                                     if entity_type == "tag" and "request_id" in transaction:
                                         update_data["request_id"] = transaction.get("request_id")
@@ -1826,6 +1854,15 @@ async def _common_add_spend_log_transaction_to_daily_transaction(
             if call_type:
                 endpoint = ROUTE_ENDPOINT_MAPPING.get(call_type, None)
 
+            cache_read_input_tokens = _extract_cache_read_tokens(usage_obj)
+            compression_saved_tokens = extract_compression_saved_tokens(_metadata)
+            savings_spend = compute_savings_spend(
+                model=payload.get("model", None),
+                custom_llm_provider=payload.get("custom_llm_provider", None),
+                compression_saved_tokens=compression_saved_tokens,
+                cache_read_input_tokens=cache_read_input_tokens,
+            )
+
             daily_transaction = BaseDailySpendTransaction(
                 date=date,
                 api_key=payload["api_key"],
@@ -1840,8 +1877,11 @@ async def _common_add_spend_log_transaction_to_daily_transaction(
                 api_requests=1,
                 successful_requests=1 if request_status == "success" else 0,
                 failed_requests=1 if request_status != "success" else 0,
-                cache_read_input_tokens=_extract_cache_read_tokens(usage_obj),
+                cache_read_input_tokens=cache_read_input_tokens,
                 cache_creation_input_tokens=_extract_cache_creation_tokens(usage_obj),
+                compression_saved_tokens=compression_saved_tokens,
+                compression_savings_spend=savings_spend.compression,
+                prompt_caching_savings_spend=savings_spend.prompt_caching,
             )
             return daily_transaction
         except Exception as e:

litellm/proxy/db/db_transaction_queue/daily_spend_update_queue.py

@@ -122,6 +122,18 @@ def get_aggregated_daily_spend_update_transactions(
                         payload.get("cache_creation_input_tokens", 0) or 0
                     ) + daily_transaction.get("cache_creation_input_tokens", 0)
 
+                    daily_transaction["compression_saved_tokens"] = (
+                        payload.get("compression_saved_tokens", 0) or 0
+                    ) + daily_transaction.get("compression_saved_tokens", 0)
+
+                    daily_transaction["compression_savings_spend"] = (
+                        payload.get("compression_savings_spend", 0) or 0
+                    ) + daily_transaction.get("compression_savings_spend", 0)
+
+                    daily_transaction["prompt_caching_savings_spend"] = (
+                        payload.get("prompt_caching_savings_spend", 0) or 0
+                    ) + daily_transaction.get("prompt_caching_savings_spend", 0)
+
                 else:
                     aggregated_daily_spend_update_transactions[_key] = deepcopy(payload)
         return aggregated_daily_spend_update_transactions

litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py

@@ -11,6 +11,7 @@
 
 import litellm
 from httpx import Response as HttpxResponse
+from litellm.proxy.spend_tracking.compression_savings import HEADROOM_GUARDRAIL_PROVIDER
 from typing_extensions import TypeGuard
 
 from litellm._logging import verbose_proxy_logger
@@ -487,7 +488,7 @@ async def apply_guardrail(
             guardrail_json_response=stats,
             request_data=request_data,
             guardrail_status="success",
-            guardrail_provider="headroom",
+            guardrail_provider=HEADROOM_GUARDRAIL_PROVIDER,
             start_time=start_time,
             end_time=end_time,
             duration=end_time - start_time,