warlock20 · warlock20 · Aug 25, 2026 · Aug 24, 2026
diff --git a/app/services/ai/config.py b/app/services/ai/config.py
@@ -471,6 +471,27 @@ def get_best_embedding_provider(self) -> Optional['EmbeddingProvider']:
 _config: Optional[AIConfig] = None
 
 
+# Symbolic tiers a prompt YAML can name instead of a concrete model id.
+# Hardcoding ids in ~20 prompt files made retiring a model a sweep across all of
+# them; naming a tier keeps the id in AIConfig, where one edit moves everything.
+MODEL_TIERS = ('default', 'quality', 'fast')
+
+
+def resolve_model_name(name: str) -> 'AIModel':
+    """Resolve a prompt's ``model`` field to an AIModel.
+
+    Accepts a tier (``default`` / ``quality`` / ``fast``), which is looked up on
+    the active config, or a concrete model id / alias.
+
+    Raises ValueError or KeyError for anything unrecognised, matching
+    ``AIModel.from_string`` so callers keep one error path.
+    """
+    key = (name or '').strip().lower()
+    if key in MODEL_TIERS:
+        return getattr(get_ai_config(), f'{key}_model')
+    return AIModel.from_string(name)
+
+
 def get_ai_config() -> AIConfig:
     """
     Get the singleton AI configuration instance.

diff --git a/app/services/ai/prompt_service.py b/app/services/ai/prompt_service.py
@@ -45,12 +45,28 @@
 import logging
 from typing import Dict, Any, Optional, List, Tuple
 from pathlib import Path
-from app.services.ai.config import AIModel, AIProvider as AIProviderEnum, get_ai_config
+from app.services.ai.config import (AIModel, AIProvider as AIProviderEnum,
+                                     get_ai_config, resolve_model_name)
 from app.models.user_ai_preferences import UserAIPreference
 
 logger = logging.getLogger(__name__)
 
 
+def _provider_for_display(prompt_data: Dict[str, Any]) -> Optional[str]:
+    """Provider name for the settings UI, derived from the prompt's model.
+
+    Prompts no longer declare a provider — the model carries it — but the
+    settings screens still want to show one.
+    """
+    model_str = prompt_data.get('model')
+    if model_str:
+        try:
+            return resolve_model_name(model_str).provider.value
+        except (ValueError, KeyError):
+            pass
+    return prompt_data.get('preferred_provider')
+
+
 class PromptService:
     """
     Central service for managing and loading LLM prompts.
@@ -206,7 +222,7 @@ def get_prompt_info(self, category: str, name: str) -> Dict[str, Any]:
             'category': category,
             'max_tokens': prompt_data.get('max_tokens', 2000),
             'temperature': prompt_data.get('temperature', 0.7),
-            'preferred_provider': prompt_data.get('preferred_provider'),
+            'preferred_provider': _provider_for_display(prompt_data),
             'model': prompt_data.get('model'),
             'required_variables': self._extract_variables(template),
             'has_system_context': 'system_context' in prompt_data,
@@ -373,10 +389,19 @@ def resolve_model_provider(
     model_str = metadata.get('model')
     if model_str:
         try:
-            model_enum = AIModel.from_string(model_str)
+            # Tier names ('quality', 'fast', ...) resolve through AIConfig, so a
+            # prompt does not have to name a concrete model id.
+            model_enum = resolve_model_name(model_str)
         except (ValueError, KeyError):
             logger.warning(f"Unknown model in prompt metadata: {model_str}")
 
+    # The provider comes from the model, which already carries it. A prompt that
+    # also declared `preferred_provider` could contradict its own model, and once
+    # prompts name a tier the YAML cannot know what that tier resolves to.
+    if model_enum is not None:
+        return model_enum, model_enum.provider
+
+    # No model named: a bare provider still selects one, using its default model.
     provider_str = metadata.get('preferred_provider')
     if provider_str:
         try:
@@ -445,7 +470,7 @@ def get_effective_model_display(
                 info = prompt_service.get_prompt_info(prompt_category, prompts[0])
                 model_str = info.get('model')
                 if model_str:
-                    model = AIModel.from_string(model_str)
+                    model = resolve_model_name(model_str)
                     return _MODEL_DISPLAY_NAMES.get(model.model_id, model.model_id)
             except (ValueError, KeyError):
                 pass

diff --git a/app/services/ai/prompts/argos/explanation.yaml b/app/services/ai/prompts/argos/explanation.yaml
@@ -6,8 +6,7 @@ description: "Generates detailed explanation for a specific Argos insight"
 version: "1.0"
 category: "argos"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 5500
 temperature: 0.5
 

diff --git a/app/services/ai/prompts/argos/relevance_scoring.yaml b/app/services/ai/prompts/argos/relevance_scoring.yaml
@@ -6,8 +6,7 @@ description: "Scores insight candidates for relevance to current research"
 version: "1.0"
 category: "argos"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 7000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/checkpoint/checkpoint_analysis.yaml b/app/services/ai/prompts/checkpoint/checkpoint_analysis.yaml
@@ -7,8 +7,7 @@ version: "1.0"
 category: "checkpoint"
 
 # AI Provider Configuration
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 1500
 temperature: 0.4
 

diff --git a/app/services/ai/prompts/companion/companion_agent.yaml b/app/services/ai/prompts/companion/companion_agent.yaml
@@ -3,8 +3,7 @@ description: "Global agentic companion — answers grounded in the user's own ac
 version: "1.0"
 category: "companion"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 # Per-turn output budget. Tool-call turns are naturally short; this headroom is for
 # the final answer when a question spans several sources and needs real synthesis.
 max_tokens: 4096

diff --git a/app/services/ai/prompts/companion/counter_evidence.yaml b/app/services/ai/prompts/companion/counter_evidence.yaml
@@ -3,8 +3,7 @@ description: "Surface disconfirming evidence and questions that challenge a rese
 version: "1.0"
 category: "companion"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 6000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/companion/knowledge_summary.yaml b/app/services/ai/prompts/companion/knowledge_summary.yaml
@@ -3,8 +3,7 @@ description: "Compress one piece of a user's research knowledge into a short, em
 version: "1.0"
 category: "companion"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "fast"
 max_tokens: 120
 temperature: 0.2
 

diff --git a/app/services/ai/prompts/companion/live_companion.yaml b/app/services/ai/prompts/companion/live_companion.yaml
@@ -3,8 +3,7 @@ description: "Answer research questions during a session, surfacing facts not op
 version: "1.0"
 category: "companion"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 6000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/companion/research_brief.yaml b/app/services/ai/prompts/companion/research_brief.yaml
@@ -3,8 +3,7 @@ description: "Generate a pre-session research brief for a research step"
 version: "1.0"
 category: "companion"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 8000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/companion/session_wrapup.yaml b/app/services/ai/prompts/companion/session_wrapup.yaml
@@ -3,8 +3,7 @@ description: "Summarize a research session: accomplishments, gaps, next steps"
 version: "1.0"
 category: "companion"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 6000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/intelligence/decision_review.yaml b/app/services/ai/prompts/intelligence/decision_review.yaml
@@ -7,7 +7,6 @@ version: "1.0"
 category: "intelligence"
 
 # AI Provider Configuration
-preferred_provider: "claude"
 model: "claude-sonnet-4-20250514"
 max_tokens: 1000
 temperature: 0.6

diff --git a/app/services/ai/prompts/intelligence/pattern_explanation.yaml b/app/services/ai/prompts/intelligence/pattern_explanation.yaml
@@ -7,7 +7,6 @@ version: "1.0"
 category: "intelligence"
 
 # AI Provider Configuration
-preferred_provider: "claude"
 model: "claude-sonnet-4-20250514"
 max_tokens: 500
 temperature: 0.5

diff --git a/app/services/ai/prompts/intelligence/research_qa.yaml b/app/services/ai/prompts/intelligence/research_qa.yaml
@@ -7,8 +7,7 @@ version: "1.0"
 category: "intelligence"
 
 # AI Provider Configuration
-preferred_provider: "gemini"  # Cost-effective for document Q&A
-model: "gemini-2.5-flash"
+model: "fast"
 max_tokens: 1000
 temperature: 0.2
 

diff --git a/app/services/ai/prompts/intelligence/text_summarization.yaml b/app/services/ai/prompts/intelligence/text_summarization.yaml
@@ -7,8 +7,7 @@ version: "1.0"
 category: "intelligence"
 
 # AI Provider Configuration
-preferred_provider: "gemini"  # Cost-effective for summarization
-model: "gemini-2.5-flash"
+model: "fast"
 max_tokens: 800
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/intelligence/thesis_analysis.yaml b/app/services/ai/prompts/intelligence/thesis_analysis.yaml
@@ -7,7 +7,6 @@ version: "1.0"
 category: "intelligence"
 
 # AI Provider Configuration
-preferred_provider: "claude"  # Can be overridden in ai_routing.yaml
 model: "claude-sonnet-4-20250514"
 max_tokens: 2000
 temperature: 0.3

diff --git a/app/services/ai/prompts/intelligence/thesis_analysis_simple.yaml b/app/services/ai/prompts/intelligence/thesis_analysis_simple.yaml
@@ -7,7 +7,6 @@ version: "1.0"
 category: "intelligence"
 
 # AI Provider Configuration
-preferred_provider: "claude"
 model: "claude-sonnet-4-20250514"
 max_tokens: 2000
 temperature: 0.3

diff --git a/app/services/ai/prompts/intelligence/warning_generation.yaml b/app/services/ai/prompts/intelligence/warning_generation.yaml
@@ -7,7 +7,6 @@ version: "1.0"
 category: "intelligence"
 
 # AI Provider Configuration
-preferred_provider: "claude"
 model: "claude-sonnet-4-20250514"
 max_tokens: 800
 temperature: 0.5

diff --git a/app/services/ai/prompts/portfolio/portfolio_complete_analysis.yaml b/app/services/ai/prompts/portfolio/portfolio_complete_analysis.yaml
@@ -6,8 +6,7 @@ description: "AI-powered analysis of portfolio data to extract behavioral insigh
 version: "1.0"
 category: "intelligence"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.5
 

diff --git a/app/services/ai/prompts/portfolio/portfolio_raw_trade_analysis.yaml b/app/services/ai/prompts/portfolio/portfolio_raw_trade_analysis.yaml
@@ -6,8 +6,7 @@ description: "Deep behavioral analysis from raw transactions with tax context, F
 version: "1.0"
 category: "portfolio"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 65000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/portfolio/sector_momentum_analysis.yaml b/app/services/ai/prompts/portfolio/sector_momentum_analysis.yaml
@@ -6,8 +6,7 @@ description: "Detect sector momentum chasing patterns using AI knowledge of hist
 version: "1.0"
 category: "portfolio"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/portfolio/tax_optimization_analysis.yaml b/app/services/ai/prompts/portfolio/tax_optimization_analysis.yaml
@@ -7,8 +7,7 @@ description: "Tax optimization analysis adapted to investor's tax residence, wit
 version: "1.0"
 category: "portfolio"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/research/bias_check.yaml b/app/services/ai/prompts/research/bias_check.yaml
@@ -6,8 +6,7 @@ description: "Analyze investment thesis for cognitive biases (Munger framework)"
 version: "1.0"
 category: "research"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.3
 

diff --git a/app/services/ai/prompts/research/challenge_mode.yaml b/app/services/ai/prompts/research/challenge_mode.yaml
@@ -6,8 +6,7 @@ description: "Generates counter-arguments and critical questions to test user's
 version: "1.0"
 category: "research"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.7
 

diff --git a/app/services/ai/prompts/research/checklist_item_analyze.yaml b/app/services/ai/prompts/research/checklist_item_analyze.yaml
@@ -6,8 +6,7 @@ description: "Analyzes financial documents and available sources to answer a spe
 version: "1.0"
 category: "research"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.4
 

diff --git a/app/services/ai/prompts/research/elaboration_mode.yaml b/app/services/ai/prompts/research/elaboration_mode.yaml
@@ -6,8 +6,7 @@ description: "Generates follow-up questions and suggests deeper areas to explore
 version: "1.0"
 category: "research"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.4
 

diff --git a/app/services/ai/prompts/research/factcheck_mode.yaml b/app/services/ai/prompts/research/factcheck_mode.yaml
@@ -6,8 +6,7 @@ description: "Verifies claims in user research with evidence-based verdicts"
 version: "2.0"
 category: "research"
 
-preferred_provider: "gemini"
-model: "gemini-3-flash-preview"
+model: "quality"
 max_tokens: 10000
 temperature: 0.2
 

diff --git a/app/services/ai/prompts/screening/screening_analysis.yaml b/app/services/ai/prompts/screening/screening_analysis.yaml
@@ -6,8 +6,7 @@ description: "Analyze kill checklist history to surface patterns and improvement
 version: "1.0"
 category: "screening"
 
-preferred_provider: "gemini"
-model: "gemini-2.5-flash"
+model: "fast"
 max_tokens: 4000
 temperature: 0.3
 

diff --git a/app/services/ai_research_assistant.py b/app/services/ai_research_assistant.py
@@ -59,7 +59,7 @@
 
 from app.services.ai import ai_service
 from app.services.ai.prompt_service import prompt_service
-from app.services.ai.config import AIProvider, AIModel
+from app.services.ai.config import AIProvider, AIModel, resolve_model_name
 
 logger = logging.getLogger(__name__)
 
@@ -292,21 +292,20 @@ def _generate_response(
             prompt_text = prompt_data['prompt']
             metadata = prompt_data['metadata']
 
-            # Convert provider and model from strings to enums
-            provider_str = metadata.get('preferred_provider')
+            # Convert the prompt's model to an enum. The provider is not read
+            # from the prompt — the model already carries it, and a separately
+            # declared provider could contradict it.
             model_str = metadata.get('model')
 
             provider_enum = None
             model_enum = None
 
-            if provider_str:
-                try:
-                    provider_enum = AIProvider(provider_str)
-                except ValueError:
-                    logger.warning(f"Invalid provider '{provider_str}', using default")
-
             if model_str:
-                model_enum = AIModel.from_string(model_str)
+                # Prompts name a tier ('quality'/'fast'), not a model id.
+                model_enum = resolve_model_name(model_str)
+                provider_enum = model_enum.provider
+
+            provider_str = provider_enum.value if provider_enum else None
 
             logger.info(
                 f"Generating {mode} response for question: {question_text[:50]}... "