refactor(agent): snapshot generation without provider mutation

nanobot/agent/loop.py

@@ -173,9 +173,15 @@ class AgentLoop:
         return self.tools.tool_names
 
     def llm_runtime(self) -> LLMRuntime:
-        """Return the current provider/model pair owned by this loop."""
+        """Capture the current provider/model settings owned by this loop."""
         self._refresh_provider_snapshot()
-        return LLMRuntime(self.provider, self.model)
+        return LLMRuntime.capture(
+            self.provider,
+            self.model,
+            context_window_tokens=self.context_window_tokens,
+            model_preset=self.model_preset,
+            snapshot_signature=self._provider_signature,
+        )
 
     _RUNTIME_CHECKPOINT_KEY = "runtime_checkpoint"
     _PENDING_USER_TURN_KEY = "pending_user_turn"
@@ -444,6 +450,8 @@ class AgentLoop:
         provider = snapshot.provider
         model = snapshot.model
         context_window_tokens = snapshot.context_window_tokens
+        if snapshot.generation is not None:
+            provider.generation = snapshot.generation
         old_model = self.model
         self.provider = provider
         self.model = model

nanobot/agent/model_presets.py

@@ -34,12 +34,12 @@ def build_static_preset_snapshot(
     name: str,
     preset: ModelPresetConfig,
 ) -> ProviderSnapshot:
-    provider.generation = preset.to_generation_settings()
     return ProviderSnapshot(
         provider=provider,
         model=preset.model,
         context_window_tokens=preset.context_window_tokens,
         signature=("model_preset", name, preset.model_dump_json()),
+        generation=preset.to_generation_settings(),
     )
 
 

nanobot/providers/factory.py

@@ -6,7 +6,7 @@ from dataclasses import dataclass
 from pathlib import Path
 
 from nanobot.config.schema import Config, InlineFallbackConfig, ModelPresetConfig, ProviderConfig
-from nanobot.providers.base import LLMProvider
+from nanobot.providers.base import GenerationSettings, LLMProvider
 from nanobot.providers.fallback_provider import FallbackProvider
 from nanobot.providers.registry import ProviderSpec, create_dynamic_spec, find_by_name
 
@@ -17,6 +17,7 @@ class ProviderSnapshot:
     model: str
     context_window_tokens: int
     signature: tuple[object, ...]
+    generation: GenerationSettings | None = None
 
 
 def _resolve_model_preset(
@@ -268,6 +269,7 @@ def build_provider_snapshot(
         model=resolved.model,
         context_window_tokens=min([resolved.context_window_tokens, *fallback_windows]),
         signature=provider_signature(config, preset=resolved),
+        generation=resolved.to_generation_settings(),
     )
 
 

nanobot/utils/llm_runtime.py

@@ -39,13 +39,18 @@ class LLMRuntime:
     ) -> LLMRuntime:
         """Capture provider defaults without retaining mutable generation state."""
         generation = provider.generation
+        defaults = GenerationSettings()
         return cls(
             provider=provider,
             model=model,
             generation=GenerationSettings(
-                temperature=generation.temperature,
+                temperature=getattr(generation, "temperature", defaults.temperature),
-                max_tokens=generation.max_tokens,
+                max_tokens=getattr(generation, "max_tokens", defaults.max_tokens),
-                reasoning_effort=generation.reasoning_effort,
+                reasoning_effort=getattr(
+                    generation,
+                    "reasoning_effort",
+                    defaults.reasoning_effort,
+                ),
             ),
             context_window_tokens=context_window_tokens,
             model_preset=model_preset,
@@ -83,6 +88,15 @@ def runtime_from_provider_snapshot(
     model_preset: str | None = None,
 ) -> LLMRuntime:
     """Convert a provider factory snapshot into the canonical runtime value."""
+    if snapshot.generation is not None:
+        return LLMRuntime(
+            provider=snapshot.provider,
+            model=snapshot.model,
+            generation=snapshot.generation,
+            context_window_tokens=snapshot.context_window_tokens,
+            model_preset=model_preset,
+            snapshot_signature=snapshot.signature,
+        )
     return LLMRuntime.capture(
         snapshot.provider,
         snapshot.model,

tests/agent/test_model_runtime_resolver.py

@@ -90,6 +90,8 @@ def test_resolver_resolves_preset_without_mutating_selected_runtime() -> None:
     assert resolved.model_preset == "fast"
     assert resolver.runtime is initial
     assert resolver.model_preset is None
+    assert initial.provider.generation == GenerationSettings(0.1, 1024, None)
+    assert resolved.generation == GenerationSettings(0.5, 512, None)
 
 
 def test_resolver_model_override_is_derived_without_default_mutation() -> None: