feat: add free distillable mode

README.md

@@ -307,10 +307,10 @@ Open-source engines for temporal AI. Render the past. Simulate the future. Score
 | **SNAG Bench** | Open Source | timepoint-snag-bench | Quality Certifier — measures Causal Resolution across renderings |
 | **Proteus** | Open Source | proteus | Settlement Layer — prediction markets that validate Rendered Futures |
 | **TDF** | Open Source | timepoint-tdf | Data Format — JSON-LD interchange across all services |
-| **Web App** | Private | timepoint-web-app | Browser client at app.timepointai.com |
-| **iPhone App** | Private | timepoint-iphone-app | iOS client — Synthetic Time Travel on mobile |
-| **Billing** | Private | timepoint-billing | Payment processing — Apple IAP + Stripe |
-| **Landing** | Private | timepoint-landing | Marketing site at timepointai.com |
+| **Web App** | — | — | Browser client at app.timepointai.com |
+| **iPhone App** | — | — | iOS client — Synthetic Time Travel on mobile |
+| **Billing** | — | — | Payment processing — Apple IAP + Stripe |
+| **Landing** | — | — | Marketing site at timepointai.com |
 
 ---
 

app/api/v1/models.py

@@ -141,6 +141,36 @@ def get_configured_models() -> list[ModelInfo]:
                 capabilities=["image_generation"],
                 pricing={"prompt": 0.00012, "completion": 0.0},
             ),
+            # NVIDIA Nemotron
+            ModelInfo(
+                id="nvidia/llama-3.3-nemotron-super-49b-v1.5",
+                name="Nemotron Super 49B v1.5",
+                provider="openrouter",
+                capabilities=["text"],
+                context_length=131072,
+            ),
+            ModelInfo(
+                id="nvidia/llama-3.1-nemotron-70b-instruct",
+                name="Nemotron 70B Instruct",
+                provider="openrouter",
+                capabilities=["text"],
+                context_length=131072,
+            ),
+            # NousResearch Hermes
+            ModelInfo(
+                id="nousresearch/hermes-4-70b",
+                name="Hermes 4 70B",
+                provider="openrouter",
+                capabilities=["text"],
+                context_length=131072,
+            ),
+            ModelInfo(
+                id="nousresearch/hermes-3-llama-3.1-405b",
+                name="Hermes 3 405B",
+                provider="openrouter",
+                capabilities=["text"],
+                context_length=131072,
+            ),
         ])
 
     return models

app/api/v1/timepoints.py

@@ -234,10 +234,14 @@ def _get_permissive_text_model() -> str:
         preference = [
             "meta-llama/llama-4-scout-17b-16e-instruct",
             "meta-llama/llama-4-maverick-17b-128e-instruct",
-            "deepseek/deepseek-chat-v3-0324",       # Fast chat model
-            "qwen/qwen3-30b-a3b",                   # Fast MoE model
+            "nvidia/llama-3.3-nemotron-super-49b-v1.5",  # Nemotron Super, fast MoE
+            "nousresearch/hermes-4-70b",             # Hermes 4 70B, strong reasoning
+            "deepseek/deepseek-chat-v3-0324",        # Fast chat model
+            "qwen/qwen3-30b-a3b",                    # Fast MoE model
+            "nvidia/nemotron-3-nano-30b-a3b",        # Nemotron Nano, very fast
+            "nousresearch/hermes-3-llama-3.1-70b",   # Hermes 3 70B fallback
             "mistralai/mistral-small-3.2-24b-instruct",
-            "qwen/qwen3-235b-a22b",                 # Large but non-thinking
+            "qwen/qwen3-235b-a22b",                  # Large but non-thinking
             "deepseek/deepseek-r1-0528",             # Thinking model — slow, last resort
         ]
         for model_id in preference:

app/config.py

@@ -54,12 +54,14 @@ class QualityPreset(str, Enum):
     - HYPER: Fastest speed with Gemini 2.0 Flash via OpenRouter
     - BALANCED: Default balance of quality and speed
     - GEMINI3: Latest Gemini 3 Flash Preview via OpenRouter (thinking model)
+    - FREE_DISTILLABLE: Free distillable models — $0 cost, outputs usable for training/distillation
     """
 
     HD = "hd"
     HYPER = "hyper"
     BALANCED = "balanced"
     GEMINI3 = "gemini3"
+    FREE_DISTILLABLE = "free_distillable"
 
 
 class Environment(str, Enum):
@@ -104,6 +106,26 @@ class VerifiedModels:
         "google/gemini-2.0-flash-001",        # Fast, handles JSON well
         "google/gemini-2.0-flash-001:free",   # Free tier (rate limited)
         "google/gemini-3-flash-preview",      # Latest thinking model, agentic workflows
+        # NVIDIA Nemotron family
+        "nvidia/llama-3.1-nemotron-70b-instruct",
+        "nvidia/llama-3.3-nemotron-super-49b-v1.5",
+        "nvidia/nemotron-3-nano-30b-a3b",
+        "nvidia/nemotron-3-nano-30b-a3b:free",
+        "nvidia/nemotron-3-super-120b-a12b:free",
+        "nvidia/nemotron-nano-12b-v2-vl",
+        "nvidia/nemotron-nano-12b-v2-vl:free",
+        "nvidia/nemotron-nano-9b-v2",
+        "nvidia/nemotron-nano-9b-v2:free",
+        # NousResearch Hermes family
+        "nousresearch/hermes-2-pro-llama-3-8b",
+        "nousresearch/hermes-3-llama-3.1-405b",
+        "nousresearch/hermes-3-llama-3.1-405b:free",
+        "nousresearch/hermes-3-llama-3.1-70b",
+        "nousresearch/hermes-4-405b",
+        "nousresearch/hermes-4-70b",
+        # OpenRouter free distillable models
+        "openrouter/hunter-alpha",
+        "openrouter/healer-alpha",
     ]
 
     # Fallback chains - ordered by preference
@@ -228,6 +250,18 @@ def is_verified_or_available(cls, model: str, provider: "ProviderType") -> bool:
         "thinking_level": "medium",  # Gemini 3 supports configurable thinking
         "image_supported": True,
     },
+    QualityPreset.FREE_DISTILLABLE: {
+        "name": "Free Distillable",
+        "description": "Free models with distillation rights — $0 cost, text-only (no image gen)",
+        "text_model": "openrouter/hunter-alpha",
+        "judge_model": "openrouter/healer-alpha",
+        "image_model": None,  # No free distillable image models available yet
+        "image_provider": None,
+        "text_provider": ProviderType.OPENROUTER,
+        "max_tokens": 4096,
+        "thinking_level": None,
+        "image_supported": False,  # Text-only mode
+    },
 }
 
 
@@ -238,6 +272,7 @@ def is_verified_or_available(cls, model: str, provider: "ProviderType") -> bool:
     QualityPreset.BALANCED: ParallelismMode.NORMAL, # Default behavior
     QualityPreset.HYPER: ParallelismMode.MAX,       # Speed focus, maximum parallelism
     QualityPreset.GEMINI3: ParallelismMode.AGGRESSIVE,  # Thinking model, moderate parallelism
+    QualityPreset.FREE_DISTILLABLE: ParallelismMode.SEQUENTIAL,  # Free models need sequential
 }
 
 # Provider rate limits (requests per minute and safe concurrent calls)
@@ -444,7 +479,7 @@ class Settings(BaseSettings):
     )
     CORS_ORIGINS: str = Field(
         default="",
-        description="Comma-separated additional CORS origins (e.g. https://your-app.up.railway.app)",
+        description="Comma-separated additional CORS origins (e.g. https://your-domain.example.com)",
     )
 
     # Share URL

app/core/llm_router.py

@@ -134,6 +134,13 @@ class ModelTier(str, Enum):
 }
 
 
+# Known free distillable models (no :free suffix but still free)
+FREE_DISTILLABLE_MODELS = {
+    "openrouter/hunter-alpha",
+    "openrouter/healer-alpha",
+}
+
+
 def is_free_model(model_id: str) -> bool:
     """Check if a model is a free tier model on OpenRouter.
 
@@ -146,7 +153,9 @@ def is_free_model(model_id: str) -> bool:
     if not model_id:
         return False
     model_lower = model_id.lower()
-    return ":free" in model_lower or "/free" in model_lower
+    if ":free" in model_lower or "/free" in model_lower:
+        return True
+    return model_lower in FREE_DISTILLABLE_MODELS
 
 
 class LLMRouter:

app/core/model_capabilities.py

@@ -404,6 +404,196 @@ class TextModelConfig:
         max_output_tokens=16384,
         notes="GPT-4o Mini via OpenRouter",
     ),
+    # NVIDIA Nemotron family (via OpenRouter)
+    "nvidia/llama-3.1-nemotron-70b-instruct": TextModelConfig(
+        model_id="nvidia/llama-3.1-nemotron-70b-instruct",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron 70B instruct, strong reasoning",
+    ),
+    "nvidia/llama-3.3-nemotron-super-49b-v1.5": TextModelConfig(
+        model_id="nvidia/llama-3.3-nemotron-super-49b-v1.5",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron Super 49B v1.5, efficient MoE",
+    ),
+    "nvidia/nemotron-3-nano-30b-a3b": TextModelConfig(
+        model_id="nvidia/nemotron-3-nano-30b-a3b",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron 3 Nano 30B, fast MoE (3B active)",
+    ),
+    "nvidia/nemotron-3-nano-30b-a3b:free": TextModelConfig(
+        model_id="nvidia/nemotron-3-nano-30b-a3b:free",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron 3 Nano 30B free tier",
+    ),
+    "nvidia/nemotron-3-super-120b-a12b:free": TextModelConfig(
+        model_id="nvidia/nemotron-3-super-120b-a12b:free",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron 3 Super 120B free tier, large MoE (12B active)",
+    ),
+    "nvidia/nemotron-nano-12b-v2-vl": TextModelConfig(
+        model_id="nvidia/nemotron-nano-12b-v2-vl",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron Nano 12B v2 with vision",
+    ),
+    "nvidia/nemotron-nano-12b-v2-vl:free": TextModelConfig(
+        model_id="nvidia/nemotron-nano-12b-v2-vl:free",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron Nano 12B v2 with vision, free tier",
+    ),
+    "nvidia/nemotron-nano-9b-v2": TextModelConfig(
+        model_id="nvidia/nemotron-nano-9b-v2",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron Nano 9B v2, compact and fast",
+    ),
+    "nvidia/nemotron-nano-9b-v2:free": TextModelConfig(
+        model_id="nvidia/nemotron-nano-9b-v2:free",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Nemotron Nano 9B v2 free tier",
+    ),
+    # NousResearch Hermes family (via OpenRouter)
+    "nousresearch/hermes-2-pro-llama-3-8b": TextModelConfig(
+        model_id="nousresearch/hermes-2-pro-llama-3-8b",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Hermes 2 Pro 8B, compact function-calling model",
+    ),
+    "nousresearch/hermes-3-llama-3.1-405b": TextModelConfig(
+        model_id="nousresearch/hermes-3-llama-3.1-405b",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Hermes 3 405B, flagship open-weight model",
+    ),
+    "nousresearch/hermes-3-llama-3.1-405b:free": TextModelConfig(
+        model_id="nousresearch/hermes-3-llama-3.1-405b:free",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Hermes 3 405B free tier",
+    ),
+    "nousresearch/hermes-3-llama-3.1-70b": TextModelConfig(
+        model_id="nousresearch/hermes-3-llama-3.1-70b",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Hermes 3 70B, strong general-purpose model",
+    ),
+    "nousresearch/hermes-4-405b": TextModelConfig(
+        model_id="nousresearch/hermes-4-405b",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Hermes 4 405B, latest flagship",
+    ),
+    "nousresearch/hermes-4-70b": TextModelConfig(
+        model_id="nousresearch/hermes-4-70b",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=8192,
+        notes="Hermes 4 70B, strong reasoning",
+    ),
+    # OpenRouter free distillable models
+    "openrouter/hunter-alpha": TextModelConfig(
+        model_id="openrouter/hunter-alpha",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=32000,
+        notes="Free distillable, 1M context, text+image input",
+    ),
+    "openrouter/healer-alpha": TextModelConfig(
+        model_id="openrouter/healer-alpha",
+        provider="openrouter",
+        supports_json_schema=False,
+        supports_json_mode=True,
+        supports_function_calling=True,
+        supports_streaming=True,
+        supports_extended_thinking=False,
+        max_output_tokens=32000,
+        notes="Free distillable, 262K context, multimodal input",
+    ),
 }
 
 # Default config for unknown models (conservative - assume JSON mode works)