sandboxy 0.0.6__tar.gz → 0.0.7__tar.gz

{sandboxy-0.0.6 → sandboxy-0.0.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sandboxy
-Version: 0.0.6
+Version: 0.0.7
 Summary: Open-source agent simulation and benchmarking platform
 Project-URL: Homepage, https://github.com/sandboxy-ai/sandboxy
 Project-URL: Repository, https://github.com/sandboxy-ai/sandboxy

{sandboxy-0.0.6 → sandboxy-0.0.7}/local-ui/src/components/ResultDisplay.tsx

@@ -2,8 +2,10 @@ import { useState } from 'react'
 import { CheckCircle, XCircle, Trophy, Clock, Zap, DollarSign, Hash, MessageSquare, ChevronDown, ChevronRight, Eye, X, AlertCircle } from 'lucide-react'
 import { RunScenarioResponse, CompareModelsResponse } from '../lib/api'
 
-export function formatCost(cost: number | null | undefined): string {
+export function formatCost(cost: number | null | undefined, isLocal?: boolean): string {
   if (cost === null || cost === undefined) return '-'
+  // Local models have cost = 0 or is_local flag
+  if (isLocal === true || cost === 0) return 'Local'
   if (cost < 0.0001) return '<$0.0001'
   if (cost < 0.01) return `$${cost.toFixed(4)}`
   return `$${cost.toFixed(3)}`
@@ -73,7 +75,7 @@ export function SingleRunResult({ result }: { result: RunScenarioResponse }) {
             <DollarSign className="w-4 h-4" />
             Cost
           </div>
-          <div className="text-2xl font-bold text-emerald-300">{formatCost(result.cost_usd)}</div>
+          <div className="text-2xl font-bold text-emerald-300">{formatCost(result.cost_usd, result.is_local)}</div>
         </div>
       </div>
 

{sandboxy-0.0.6 → sandboxy-0.0.7}/local-ui/src/lib/api.ts

@@ -110,6 +110,8 @@ export interface RunScenarioResponse {
   output_tokens: number
   cost_usd: number | null
   error: string | null
+  is_local?: boolean
+  provider_name?: string
 }
 
 export interface CompareModelsRequest {

{sandboxy-0.0.6 → sandboxy-0.0.7}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "sandboxy"
-version = "0.0.6"
+version = "0.0.7"
 description = "Open-source agent simulation and benchmarking platform"
 readme = "README.md"
 license = "Apache-2.0"

{sandboxy-0.0.6 → sandboxy-0.0.7}/sandboxy/api/app.py

@@ -36,6 +36,19 @@ def create_local_app(
         root_dir: Working directory for scenarios/tools/agents.
         local_ui_path: Path to local UI static files.
     """
+    # Load .env from root_dir to pick up API keys from the project directory
+    # This is important when root_dir differs from cwd
+    env_file = root_dir / ".env"
+    if env_file.exists():
+        load_dotenv(env_file, override=True)
+        logger.info(f"Loaded environment from {env_file}")
+
+        # Reset provider registry so it picks up the newly loaded API keys
+        from sandboxy.providers.registry import reset_registry
+
+        reset_registry()
+        logger.info("Provider registry reset after loading project .env")
+
     from sandboxy.local.context import LocalContext, set_local_context
 
     ctx = LocalContext(root_dir=root_dir)

{sandboxy-0.0.6 → sandboxy-0.0.7}/sandboxy/api/routes/local.py

@@ -493,7 +493,10 @@ async def run_scenario(request: RunScenarioRequest) -> RunScenarioResponse:
                     experiment_name=mlflow_config.experiment,
                 )
 
-            with mlflow_run_context(mlflow_config, run_name=request.model) as run_id:
+            # Short run name - just the model (scenario is in experiment name)
+            run_name = request.model.split("/")[-1] if "/" in request.model else request.model
+
+            with mlflow_run_context(mlflow_config, run_name=run_name) as run_id:
                 result = await runner.run(
                     scenario=spec,
                     model=request.model,
@@ -633,12 +636,17 @@ async def compare_models(request: CompareModelsRequest) -> CompareModelsResponse
                         scenario_name=spec.name,
                     )
                     exporter = MLflowExporter(config)
+
+                    # Short run name - just the model
+                    run_name = result.model.split("/")[-1] if "/" in result.model else result.model
+
                     exporter.export(
                         result=result.to_dict(),
                         scenario_path=scenario_path,
                         scenario_name=spec.name,
                         scenario_id=spec.id,
                         agent_name=result.model,
+                        run_name=run_name,
                     )
             except ImportError:
                 logger.warning("MLflow not installed, skipping export")
@@ -1513,7 +1521,9 @@ async def run_with_dataset(request: RunDatasetRequest) -> RunDatasetResponse:
         if mlflow_config and mlflow_config.enabled:
             from sandboxy.mlflow import mlflow_run_context
 
-            run_name = f"{request.model}-{request.dataset_id}"
+            # Short run name - model + dataset
+            model_short = request.model.split("/")[-1] if "/" in request.model else request.model
+            run_name = f"{model_short} ({request.dataset_id})"
             with mlflow_run_context(mlflow_config, run_name=run_name) as run_id:
                 result = await run_dataset_benchmark()
 

{sandboxy-0.0.6 → sandboxy-0.0.7}/sandboxy/cli/main.py

@@ -833,7 +833,9 @@ def scenario(
                 )
 
             # Start run, execute scenario, then log metrics - all connected
-            with mlflow_run_context(mlflow_config, run_name=model_id) as run_id:
+            # Short run name - just the model
+            run_name = model_id.split("/")[-1] if "/" in model_id else model_id
+            with mlflow_run_context(mlflow_config, run_name=run_name) as run_id:
                 runner = ScenarioRunner(scenario=spec, agent=agent)
                 result = runner.run(max_turns=max_turns)
 

{sandboxy-0.0.6 → sandboxy-0.0.7}/sandboxy/providers/__init__.py

@@ -40,6 +40,7 @@ from sandboxy.providers.registry import (
     get_provider,
     get_registry,
     reload_local_providers,
+    reset_registry,
 )
 
 __all__ = [
@@ -53,6 +54,7 @@ __all__ = [
     "get_provider",
     "get_registry",
     "reload_local_providers",
+    "reset_registry",
     # Local provider
     "LocalProvider",
     "LocalProviderConnectionError",

{sandboxy-0.0.6 → sandboxy-0.0.7}/sandboxy/providers/registry.py

@@ -154,23 +154,34 @@ class ProviderRegistry:
         if "/" in model_id:
             provider_name, model_name = model_id.split("/", 1)
 
-            # Check for local provider first (e.g., "ollama/llama3")
+            # Priority 1: Check for LOCAL provider with matching name
+            # Local providers take precedence over cloud providers
             if provider_name in self.providers:
                 provider = self.providers[provider_name]
-                # Verify it's a local provider or supports the model
-                if hasattr(provider, "config") or provider.supports_model(model_id):
+                # Only use if it's a local provider (has config attribute)
+                if hasattr(provider, "config"):
                     return provider
 
-            # OpenRouter format (e.g., "openai/gpt-4o")
+            # Priority 2: OpenRouter for provider/model format (e.g., "openai/gpt-4o")
             if "openrouter" in self.providers:
                 return self.providers["openrouter"]
 
-            # Fallback to direct provider if prefix matches
+            # Priority 3: Fallback to direct cloud provider if available
             if provider_name == "openai" and "openai" in self.providers:
                 return self.providers["openai"]
             if provider_name == "anthropic" and "anthropic" in self.providers:
                 return self.providers["anthropic"]
 
+            # No valid provider found for prefixed model - raise clear error
+            available = list(self.providers.keys())
+            raise ProviderError(
+                f"No provider available for model '{model_id}'. "
+                f"The '{provider_name}/' prefix requires OpenRouter (set OPENROUTER_API_KEY) "
+                f"or a local provider named '{provider_name}'. "
+                f"Available providers: {available}",
+                provider="registry",
+            )
+
         # No prefix - use direct providers
         model_lower = model_id.lower()
 
@@ -280,6 +291,16 @@ def get_registry() -> ProviderRegistry:
     return _registry
 
 
+def reset_registry() -> None:
+    """Reset the global provider registry.
+
+    Forces re-initialization on next get_registry() call.
+    Useful after loading new environment variables.
+    """
+    global _registry
+    _registry = None
+
+
 def get_provider(model_id: str) -> BaseProvider:
     """Get a provider for a model (convenience function).