Chore/model bump

app/appai/services/agents/deck_constructor.py

@@ -104,6 +104,7 @@
   - Is the card affordable in terms of mana cost and mana colors that the deck can produce? If not, is it worth adjusting the mana base to accommodate the card, or should it be avoided in favor of a more affordable option?
 - Consider the balance of the deck, including the mana curve, the mix of card types, and the overall consistency of the deck.
 - Do not assume names of specific cards, beyond basic lands. The legal sets are constantly changing, so you cannot rely on prior knowledge of specific cards being available.
+- When starting off, try to launch multiple searches for cards that will cover different aspects of the deck's strategy. By running searches in parallel, you can save time for the user.
 
 ## General flow:
 Remember, a deck must have at least 60 cards, and no more than 4 copies of any individual card (other than basic lands).
@@ -194,6 +195,7 @@ async def run_deck_constructor_agent(
     agent = Agent(
         system_prompt=DECK_CONSTRUCTION_SYSTEM_PROMPT,
         model=TOOL_MODEL_THINKING,
+        model_settings={'thinking': 'high'},
         deps_type=DeckBuildingDeps,
         tools=[
             subagent_memory_search,
@@ -358,6 +360,7 @@ async def run_card_classifier_agent(deck_id: UUID, deck_description: str) -> Non
     agent = Agent(
         system_prompt=CARD_CLASSIFIER_SYSTEM_PROMPT,
         model=TOOL_MODEL_BASIC,
+        model_settings={'thinking': False},
         tools=[],
         instrument=True,
         retries=0,
@@ -448,6 +451,7 @@ async def run_card_replacement_agent(
     agent = Agent(
         system_prompt=CARD_REPLACEMENT_SYSTEM_PROMPT,
         model=TOOL_MODEL_BASIC,
+        model_settings={'thinking': False},
         tools=[],
         instrument=True,
         retries=0,

app/appai/services/agents/deck_theme.py

@@ -80,6 +80,7 @@ def get_daily_deck_theme() -> NewTheme:
     agent = Agent(
         model=TOOL_MODEL_BASIC,
         system_prompt=DECK_THEME_PROMPT,
+        model_settings={'thinking': False},
         instrument=True,
         output_type=NewTheme,
         tools=[find_similar_themes],

app/appai/services/agents/filter_constructor.py

@@ -220,6 +220,7 @@ async def filter_constructor(query: str) -> Filter:
     agent = Agent(
         model=TOOL_MODEL_BASIC,
         system_prompt=FILTER_CONSTRUCTION_PROMPT,
+        model_settings={'thinking': False},
         output_retries=10,
         output_type=Filter,
         instrument=True,

app/appai/services/agents/guardrails.py

@@ -87,6 +87,7 @@ def guardrail_agent(user_request: str, context: str) -> RelevancyScore:
 
     agent = Agent(
         model=TEXT_MODEL,
+        model_settings={'thinking': False},
         system_prompt=get_guardrail_system_prompt(context),
         output_retries=5,
         output_type=RelevancyScore,

app/appai/services/agents/tools/memory_tools.py

@@ -133,7 +133,8 @@ async def write_memory(ctx: RunContext[DeckBuildingDeps], content: str, related_
         None: This tool does not return any output, it simply records the memory for future reference.
     """
     if len(related_card_uuids) > 10:
-        raise ValueError("A maximum of 10 related card UUIDs can be included in a memory.")
+        logfire.warning("Memory writing tool called with too many related_card_uuids.", count=len(related_card_uuids))
+        raise ModelRetry("A maximum of 10 related card UUIDs can be included in a memory.")
     if len(related_card_uuids) > 0:
         try:
             await _check_related_card_uuids(related_card_uuids)
@@ -147,7 +148,8 @@ async def write_memory(ctx: RunContext[DeckBuildingDeps], content: str, related_
     agent = Agent(
         model=TOOL_MODEL_BASIC,
         system_prompt=MEMORY_WRITING_AGENT_PROMPT,
-        output_type=Memory | None,  # type: ignore [arg-type]
+        model_settings={'thinking': False},
+        output_type=Memory | None,  # type: ignore [arg-type, call-overload]
         retries=10,
         output_retries=10,
         instrument=True,
@@ -394,6 +396,7 @@ async def subagent_memory_search(ctx: RunContext[DeckBuildingDeps], query: str)
     agent = Agent(
         model=TOOL_MODEL_BASIC,
         system_prompt=MEMORY_SEARCH_PROMPT,
+        model_settings={'thinking': False},
         instrument=True,
         tools=[semantic_memory_search, card_memory_search],
         output_type=MemorySummary,

pyproject.toml

@@ -23,7 +23,6 @@ dependencies = [
   "jq>=1.11.0",
   "tqdm>=4.67.1",
   "kombu>=5.6.2",
-  "pydantic-ai>=1.58.0",
   "pydantic>=2.12.5",
   "pydantic-settings>=2.12.0",
   "qdrant-client>=1.16.2",
@@ -40,6 +39,7 @@ dependencies = [
   "authlib>=1.6.9",
   "langchain-core>=1.2.22",
   "pyasn1>=0.6.3",
+  "pydantic-ai>=1.87.0",
 ]
 
 [dependency-groups]