Fix: load config for summarizer model from defaults instead of agent's config (#9568)

* load default settings instead of loading from agent for summarizer config

* update tests to allow use of get_llm_config_from_handle

* remove nit comment

---------

Co-authored-by: Amy Guan <amy@letta.com>

letta/services/summarizer/compact.py

@@ -73,48 +73,21 @@ async def build_summarizer_llm_config(
         return agent_llm_config
 
     try:
-        # Parse provider/model from the handle, falling back to the agent's
-        # provider type when only a model name is given.
-        if "/" in summarizer_config.model:
-            provider_name, model_name = summarizer_config.model.split("/", 1)
-        else:
-            provider_name = agent_llm_config.provider_name
-            model_name = summarizer_config.model
+        # Load default config for the summarizer model handle, using the agent's context window
+        from letta.services.provider_manager import ProviderManager
 
-        # Start from the agent's config and override model + provider_name + handle
-        # Check if the summarizer's provider matches the agent's provider
-        # If they match, we can safely use the agent's config as a base
-        # If they don't match, we need to load the default config for the new provider
-
-        provider_matches = False
+        provider_manager = ProviderManager()
         try:
-            # Check if provider_name is a valid ProviderType that matches agent's endpoint type
-            provider_type = ProviderType(provider_name)
-            provider_matches = provider_type.value == agent_llm_config.model_endpoint_type
-        except ValueError:
-            # provider_name is a custom label - check if it matches agent's provider_name
-            provider_matches = provider_name == agent_llm_config.provider_name
-
-        if provider_matches:
-            # Same provider - use agent's config as base and override model/handle
-            base = agent_llm_config.model_copy()
-            base.model = model_name
-            base.handle = summarizer_config.model
-        else:
-            # Different provider - load default config for this handle
-            from letta.services.provider_manager import ProviderManager
-
-            provider_manager = ProviderManager()
-            try:
-                base = await provider_manager.get_llm_config_from_handle(
-                    handle=summarizer_config.model,
-                    actor=actor,
-                )
-            except Exception as e:
-                logger.warning(
-                    f"Failed to load LLM config for summarizer handle '{summarizer_config.model}': {e}. Falling back to agent's LLM config."
-                )
-                return agent_llm_config
+            # automatically sets the context window to the max available for the summarizer model
+            base = await provider_manager.get_llm_config_from_handle(
+                handle=summarizer_config.model,
+                actor=actor,
+            )
+        except Exception as e:
+            logger.warning(
+                f"Failed to load LLM config for summarizer handle '{summarizer_config.model}': {e}. Falling back to agent's LLM config."
+            )
+            return agent_llm_config
 
         # If explicit model_settings are provided for the summarizer, apply
         # them just like server.create_agent_async does for agents.

letta/services/summarizer/summarizer_config.py

@@ -11,9 +11,9 @@ from letta.settings import summarizer_settings
 def get_default_summarizer_model(provider_type: ProviderType) -> str | None:
     """Get default model for summarization for given provider type."""
     summarizer_defaults = {
-        ProviderType.anthropic: "anthropic/claude-haiku-4-5",
+        ProviderType.anthropic: "anthropic/claude-haiku-4-5-20251001",
         ProviderType.openai: "openai/gpt-5-mini",
-        ProviderType.google_ai: "google_ai/gemini-2.0-flash",
+        ProviderType.google_ai: "google_ai/gemini-2.5-flash",
     }
     return summarizer_defaults.get(provider_type)
 

tests/managers/test_agent_manager.py

@@ -242,7 +242,7 @@ async def test_create_agent_with_model_handle_uses_correct_llm_config(server: Sy
 
 
 @pytest.mark.asyncio
-async def test_compaction_settings_model_uses_separate_llm_config_for_summarization(default_user):
+async def test_compaction_settings_model_uses_separate_llm_config_for_summarization(server: SyncServer, default_user):
     """When compaction_settings.model differs from the agent model, use a separate llm_config.
 
     This test exercises the summarization helpers directly to avoid external
@@ -257,6 +257,8 @@ async def test_compaction_settings_model_uses_separate_llm_config_for_summarizat
     from letta.schemas.model import OpenAIModelSettings, OpenAIReasoning
     from letta.services.summarizer.compact import build_summarizer_llm_config
 
+    await server.init_async(init_with_default_org_and_user=True)
+
     # Base agent LLM config
     base_llm_config = LLMConfig.default_config("gpt-4o-mini")
     assert base_llm_config.model == "gpt-4o-mini"
@@ -382,7 +384,7 @@ async def test_create_agent_sets_default_compaction_model_anthropic(server: Sync
 
     # Should have default haiku model set
     assert agent.compaction_settings is not None
-    assert agent.compaction_settings.model == "anthropic/claude-haiku-4-5"
+    assert agent.compaction_settings.model == "anthropic/claude-haiku-4-5-20251001"
 
 
 @pytest.mark.asyncio