feat(server): add server side cache

evaluators/contrib/galileo/src/agent_control_evaluator_galileo/records/factory.py

@@ -3,7 +3,7 @@
 from __future__ import annotations
 
 from collections.abc import Mapping
-from typing import Any
+from typing import Any, cast
 
 from agent_control_models import Step
 from galileo_core.schemas.logging.session import Session
@@ -156,21 +156,21 @@ def record_from_step(
             **common,
             input=GalileoRecordNormalizer.tool_input(input_value),
             output=GalileoRecordNormalizer.tool_output(output_value),
-            spans=_span_children(child_records, parent_type=record_type),
+            spans=cast(Any, _span_children(child_records, parent_type=record_type)),
         )
     if record_type == "retriever":
         return RetrieverSpan(
             **common,
             input=GalileoRecordNormalizer.retriever_input(input_value),
             output=GalileoRecordNormalizer.retriever_output(output_value),
-            spans=_span_children(child_records, parent_type=record_type),
+            spans=cast(Any, _span_children(child_records, parent_type=record_type)),
         )
     if record_type == "trace":
         return Trace(
             **common,
             input=GalileoRecordNormalizer.trace_input(input_value),
             output=GalileoRecordNormalizer.trace_output(output_value),
-            spans=_span_children(child_records, parent_type=record_type),
+            spans=cast(Any, _span_children(child_records, parent_type=record_type)),
         )
     traces = [child for child in child_records if isinstance(child, Trace)]
     if len(traces) != len(child_records):

server/pyproject.toml

@@ -34,6 +34,7 @@ license = {text = "Apache-2.0"}
 
 [project.optional-dependencies]
 galileo = ["agent-control-evaluator-galileo>=7.5.0"]
+redis = ["redis>=5.2.1"]
 binary = ["psycopg[binary]>=3.1"]
 c = ["psycopg[c]>=3.1"]
 

server/src/agent_control_server/config.py

@@ -3,7 +3,7 @@
 import logging
 import secrets
 from functools import cached_property
-from typing import Any
+from typing import Any, Literal
 
 from agent_control_models import JSONObject
 from agent_control_telemetry import DEFAULT_CONTROL_EVENT_SINK_NAME
@@ -191,6 +191,21 @@ class Settings(BaseSettings):
         ),
     )
 
+    control_cache_backend: Literal["postgres", "redis"] = Field(
+        default="postgres", validation_alias=AliasChoices("AGENT_CONTROL_CONTROL_CACHE_BACKEND")
+    )
+    redis_url: str | None = Field(
+        default=None, validation_alias=AliasChoices("AGENT_CONTROL_REDIS_URL")
+    )
+    redis_mode: Literal["standalone", "cluster"] = Field(
+        default="standalone", validation_alias=AliasChoices("AGENT_CONTROL_REDIS_MODE")
+    )
+    control_cache_ttl_seconds: int = Field(
+        default=60,
+        ge=1,
+        validation_alias=AliasChoices("AGENT_CONTROL_CONTROL_CACHE_TTL_SECONDS"),
+    )
+
     # Prometheus metrics settings
     prometheus_metrics_prefix: str = _env_alias_field(
         "agent_control_server",