sandboxy 0.0.3__tar.gz → 0.0.4__tar.gz

{sandboxy-0.0.3 → sandboxy-0.0.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sandboxy
-Version: 0.0.3
+Version: 0.0.4
 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
@@ -39,6 +39,8 @@ Requires-Dist: pytest-xdist>=3.5.0; extra == 'dev'
 Requires-Dist: pytest>=8.0; extra == 'dev'
 Requires-Dist: respx>=0.21.0; extra == 'dev'
 Requires-Dist: ruff>=0.1; extra == 'dev'
+Provides-Extra: mlflow
+Requires-Dist: mlflow>=3.0; extra == 'mlflow'
 Description-Content-Type: text/markdown
 
 # Sandboxy
@@ -204,6 +206,39 @@ sandboxy list-models --search claude
 sandboxy list-models --free
 ```
 
+## MLflow Integration
+
+Export scenario run results to MLflow for experiment tracking and model comparison.
+
+```bash
+# Install with MLflow support
+pip install sandboxy[mlflow]
+
+# Export run to MLflow
+sandboxy scenario scenarios/test.yml -m openai/gpt-4o --mlflow-export
+
+# Custom experiment name
+sandboxy scenario scenarios/test.yml -m gpt-4o --mlflow-export --mlflow-experiment "my-evals"
+```
+
+Or enable in scenario YAML:
+
+```yaml
+id: my-scenario
+name: "My Test"
+
+mlflow:
+  enabled: true
+  experiment: "agent-evals"
+  tags:
+    team: "support"
+
+system_prompt: |
+  ...
+```
+
+See `MLFLOW_TRACKING_URI` env variable to configure the MLflow server.
+
 ## Configuration
 
 Environment variables (in `~/.sandboxy/.env` or project `.env`):
@@ -213,6 +248,7 @@ Environment variables (in `~/.sandboxy/.env` or project `.env`):
 | `OPENROUTER_API_KEY` | OpenRouter API key (400+ models) |
 | `OPENAI_API_KEY` | Direct OpenAI access |
 | `ANTHROPIC_API_KEY` | Direct Anthropic access |
+| `MLFLOW_TRACKING_URI` | MLflow tracking server URI |
 
 ## Project Structure
 

{sandboxy-0.0.3 → sandboxy-0.0.4}/README.md

@@ -161,6 +161,39 @@ sandboxy list-models --search claude
 sandboxy list-models --free
 ```
 
+## MLflow Integration
+
+Export scenario run results to MLflow for experiment tracking and model comparison.
+
+```bash
+# Install with MLflow support
+pip install sandboxy[mlflow]
+
+# Export run to MLflow
+sandboxy scenario scenarios/test.yml -m openai/gpt-4o --mlflow-export
+
+# Custom experiment name
+sandboxy scenario scenarios/test.yml -m gpt-4o --mlflow-export --mlflow-experiment "my-evals"
+```
+
+Or enable in scenario YAML:
+
+```yaml
+id: my-scenario
+name: "My Test"
+
+mlflow:
+  enabled: true
+  experiment: "agent-evals"
+  tags:
+    team: "support"
+
+system_prompt: |
+  ...
+```
+
+See `MLFLOW_TRACKING_URI` env variable to configure the MLflow server.
+
 ## Configuration
 
 Environment variables (in `~/.sandboxy/.env` or project `.env`):
@@ -170,6 +203,7 @@ Environment variables (in `~/.sandboxy/.env` or project `.env`):
 | `OPENROUTER_API_KEY` | OpenRouter API key (400+ models) |
 | `OPENAI_API_KEY` | Direct OpenAI access |
 | `ANTHROPIC_API_KEY` | Direct Anthropic access |
+| `MLFLOW_TRACKING_URI` | MLflow tracking server URI |
 
 ## Project Structure
 

{sandboxy-0.0.3 → sandboxy-0.0.4}/local-ui/src/hooks/useScenarioRun.ts

@@ -7,13 +7,20 @@ import { api, RunScenarioResponse, CompareModelsResponse } from '../lib/api'
 
 export type RunState = 'idle' | 'running' | 'completed' | 'error'
 
+export interface MlflowOptions {
+  enabled: boolean
+  trackingUri?: string
+  experiment?: string
+  tracing?: boolean
+}
+
 export interface UseScenarioRunResult {
   state: RunState
   result: RunScenarioResponse | null
   comparison: CompareModelsResponse | null
   error: string | null
-  runScenario: (scenarioId: string, model: string, variables?: Record<string, unknown>) => Promise<void>
-  compareModels: (scenarioId: string, models: string[], runsPerModel?: number, variables?: Record<string, unknown>) => Promise<void>
+  runScenario: (scenarioId: string, model: string, variables?: Record<string, unknown>, mlflow?: MlflowOptions) => Promise<void>
+  compareModels: (scenarioId: string, models: string[], runsPerModel?: number, variables?: Record<string, unknown>, mlflow?: MlflowOptions) => Promise<void>
   reset: () => void
 }
 
@@ -33,7 +40,8 @@ export function useScenarioRun(): UseScenarioRunResult {
   const runScenario = useCallback(async (
     scenarioId: string,
     model: string,
-    variables?: Record<string, unknown>
+    variables?: Record<string, unknown>,
+    mlflow?: MlflowOptions
   ) => {
     reset()
     setState('running')
@@ -43,6 +51,10 @@ export function useScenarioRun(): UseScenarioRunResult {
         scenario_id: scenarioId,
         model,
         variables,
+        mlflow_export: mlflow?.enabled,
+        mlflow_tracking_uri: mlflow?.trackingUri,
+        mlflow_experiment: mlflow?.experiment,
+        mlflow_tracing: mlflow?.tracing,
       })
 
       if (response.error) {
@@ -62,7 +74,8 @@ export function useScenarioRun(): UseScenarioRunResult {
     scenarioId: string,
     models: string[],
     runsPerModel: number = 1,
-    variables?: Record<string, unknown>
+    variables?: Record<string, unknown>,
+    mlflow?: MlflowOptions
   ) => {
     reset()
     setState('running')
@@ -73,6 +86,10 @@ export function useScenarioRun(): UseScenarioRunResult {
         models,
         runs_per_model: runsPerModel,
         variables,
+        mlflow_export: mlflow?.enabled,
+        mlflow_tracking_uri: mlflow?.trackingUri,
+        mlflow_experiment: mlflow?.experiment,
+        mlflow_tracing: mlflow?.tracing,
       })
 
       setState('completed')

{sandboxy-0.0.3 → sandboxy-0.0.4}/local-ui/src/lib/api.ts

@@ -53,6 +53,10 @@ export interface RunScenarioRequest {
   max_turns?: number
   max_tokens?: number
   temperature?: number
+  mlflow_export?: boolean
+  mlflow_tracking_uri?: string
+  mlflow_experiment?: string
+  mlflow_tracing?: boolean
 }
 
 export interface HistoryMessage {
@@ -112,6 +116,10 @@ export interface CompareModelsRequest {
   runs_per_model?: number
   variables?: Record<string, unknown>
   max_turns?: number
+  mlflow_export?: boolean
+  mlflow_tracking_uri?: string
+  mlflow_experiment?: string
+  mlflow_tracing?: boolean
 }
 
 export interface ModelStats {
@@ -205,6 +213,8 @@ export interface RunDatasetRequest {
   max_tokens?: number
   temperature?: number
   parallel?: number
+  mlflow_enabled?: boolean
+  mlflow_experiment?: string
 }
 
 export interface CaseResultInfo {

{sandboxy-0.0.3 → sandboxy-0.0.4}/local-ui/src/pages/RunPage.tsx

@@ -1,8 +1,8 @@
 import { useState, useEffect } from 'react'
 import { useParams, useSearchParams, Link } from 'react-router-dom'
-import { ArrowLeft, Play, Loader2, XCircle, Edit, Settings, Database, Check, X } from 'lucide-react'
+import { ArrowLeft, Play, Loader2, XCircle, Edit, Settings, Database, Check, X, ChevronDown, ChevronRight, ExternalLink } from 'lucide-react'
 import { api, ScenarioDetail, ModelInfo, VariableInfo, DatasetInfo, RunDatasetResponse, LocalFileInfo } from '../lib/api'
-import { useScenarioRun } from '../hooks/useScenarioRun'
+import { useScenarioRun, MlflowOptions } from '../hooks/useScenarioRun'
 import { SingleRunResult, ComparisonResult } from '../components/ResultDisplay'
 import { ModelSelector, MultiModelSelector } from '../components/ModelSelector'
 
@@ -33,6 +33,13 @@ export default function RunPage() {
   const [datasetRunning, setDatasetRunning] = useState(false)
   const [parallel, setParallel] = useState(parallelFromUrl)
 
+  // MLflow state
+  const [mlflowEnabled, setMlflowEnabled] = useState(false)
+  const [mlflowExpanded, setMlflowExpanded] = useState(false)
+  const [mlflowTrackingUri, setMlflowTrackingUri] = useState('')
+  const [mlflowExperiment, setMlflowExperiment] = useState('')
+  const [mlflowTracing, setMlflowTracing] = useState(true)
+
   const { state, result, comparison, error: runError, runScenario, compareModels } = useScenarioRun()
 
   useEffect(() => {
@@ -93,6 +100,13 @@ export default function RunPage() {
     const sid = selectedScenarioId || scenarioId
     if (!sid) return
 
+    const mlflowOptions: MlflowOptions | undefined = mlflowEnabled ? {
+      enabled: true,
+      trackingUri: mlflowTrackingUri || undefined,
+      experiment: mlflowExperiment || undefined,
+      tracing: mlflowTracing,
+    } : undefined
+
     if (runMode === 'dataset') {
       if (!selectedDataset || !selectedModel) return
       setDatasetRunning(true)
@@ -103,6 +117,8 @@ export default function RunPage() {
           dataset_id: selectedDataset,
           model: selectedModel,
           parallel,
+          mlflow_enabled: mlflowEnabled,
+          mlflow_experiment: mlflowExperiment || undefined,
         })
         setDatasetResult(result)
       } catch (err) {
@@ -112,10 +128,10 @@ export default function RunPage() {
       }
     } else if (runMode === 'single') {
       if (!selectedModel) return
-      await runScenario(sid, selectedModel, variables)
+      await runScenario(sid, selectedModel, variables, mlflowOptions)
     } else {
       if (selectedModels.length === 0) return
-      await compareModels(sid, selectedModels, runsPerModel, variables)
+      await compareModels(sid, selectedModels, runsPerModel, variables, mlflowOptions)
     }
   }
 
@@ -360,6 +376,96 @@ export default function RunPage() {
           </div>
         )}
 
+        {/* MLflow Section */}
+        <div className="mb-6 p-4 panel-subtle">
+          <button
+            onClick={() => setMlflowExpanded(!mlflowExpanded)}
+            className="flex items-center gap-2 w-full text-left"
+          >
+            {mlflowExpanded ? <ChevronDown size={18} /> : <ChevronRight size={18} />}
+            <span className="font-medium text-slate-100">MLflow Tracking</span>
+            {mlflowEnabled && (
+              <span className="ml-2 px-2 py-0.5 text-xs bg-green-500/20 text-green-400 rounded">
+                Enabled
+              </span>
+            )}
+          </button>
+
+          {mlflowExpanded && (
+            <div className="mt-4 space-y-4">
+              {/* Enable Toggle */}
+              <label className="flex items-center gap-3 cursor-pointer">
+                <input
+                  type="checkbox"
+                  checked={mlflowEnabled}
+                  onChange={(e) => setMlflowEnabled(e.target.checked)}
+                  disabled={state === 'running' || datasetRunning}
+                  className="w-4 h-4 rounded border-slate-600 text-orange-400 focus:ring-orange-400"
+                />
+                <span className="text-slate-200">Enable MLflow tracking</span>
+              </label>
+
+              {mlflowEnabled && (
+                <>
+                  {/* Tracking URI - only for non-dataset runs (dataset uses env var) */}
+                  {runMode !== 'dataset' && (
+                    <div>
+                      <label className="block text-sm font-medium text-slate-400 mb-1">
+                        Tracking URI
+                      </label>
+                      <input
+                        type="text"
+                        value={mlflowTrackingUri}
+                        onChange={(e) => setMlflowTrackingUri(e.target.value)}
+                        disabled={state === 'running'}
+                        placeholder="http://127.0.0.1:5000 (uses MLFLOW_TRACKING_URI if empty)"
+                        className="w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400"
+                      />
+                    </div>
+                  )}
+
+                  {/* Experiment Name */}
+                  <div>
+                    <label className="block text-sm font-medium text-slate-400 mb-1">
+                      Experiment Name
+                    </label>
+                    <input
+                      type="text"
+                      value={mlflowExperiment}
+                      onChange={(e) => setMlflowExperiment(e.target.value)}
+                      disabled={state === 'running' || datasetRunning}
+                      placeholder={runMode === 'dataset' ? `${scenario?.name || 'scenario'}-dataset` : (scenario?.name || 'Defaults to scenario name')}
+                      className="w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400"
+                    />
+                  </div>
+
+                  {/* Tracing Toggle - only for non-dataset runs */}
+                  {runMode !== 'dataset' && (
+                    <label className="flex items-center gap-3 cursor-pointer">
+                      <input
+                        type="checkbox"
+                        checked={mlflowTracing}
+                        onChange={(e) => setMlflowTracing(e.target.checked)}
+                        disabled={state === 'running'}
+                        className="w-4 h-4 rounded border-slate-600 text-orange-400 focus:ring-orange-400"
+                      />
+                      <div>
+                        <span className="text-slate-200">Enable LLM Tracing</span>
+                        <p className="text-xs text-slate-500">Capture detailed traces of each LLM call</p>
+                      </div>
+                    </label>
+                  )}
+
+                  <p className="text-xs text-slate-500 flex items-center gap-1">
+                    <ExternalLink size={12} />
+                    View results at your MLflow server after the run completes
+                  </p>
+                </>
+              )}
+            </div>
+          )}
+        </div>
+
         <button
           onClick={handleRun}
           disabled={

{sandboxy-0.0.3 → sandboxy-0.0.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "sandboxy"
-version = "0.0.3"
+version = "0.0.4"
 description = "Open-source agent simulation and benchmarking platform"
 readme = "README.md"
 license = "Apache-2.0"
@@ -41,6 +41,9 @@ dependencies = [
 ]
 
 [project.optional-dependencies]
+mlflow = [
+    "mlflow>=3.0",
+]
 dev = [
     "pytest>=8.0",
     "pytest-cov>=4.0",
@@ -120,6 +123,8 @@ ignore = [
 # S307: eval usage with safety measures (safe_builtins, simpleeval)
 "sandboxy/tools/yaml_tools.py" = ["S307"]
 "sandboxy/core/safe_eval.py" = ["S307"]
+# S603, S607, S110: git subprocess call is safe (hardcoded command, known input)
+"sandboxy/mlflow/tags.py" = ["S603", "S607", "S110"]
 
 [tool.ruff.lint.pydocstyle]
 convention = "google"

{sandboxy-0.0.3 → sandboxy-0.0.4}/sandboxy/api/routes/local.py

@@ -379,6 +379,10 @@ class RunScenarioRequest(BaseModel):
     max_turns: int = 20
     max_tokens: int = 1024
     temperature: float = 0.7
+    mlflow_export: bool = False
+    mlflow_tracking_uri: str | None = None
+    mlflow_experiment: str | None = None
+    mlflow_tracing: bool = True
 
 
 class RunScenarioResponse(BaseModel):
@@ -393,6 +397,9 @@ class RunScenarioResponse(BaseModel):
     final_state: dict[str, Any]
     evaluation: dict[str, Any] | None
     latency_ms: int
+    input_tokens: int = 0
+    output_tokens: int = 0
+    cost_usd: float | None = None
     error: str | None
 
 
@@ -404,6 +411,10 @@ class CompareModelsRequest(BaseModel):
     runs_per_model: int = 1
     variables: dict[str, Any] = Field(default_factory=dict)
     max_turns: int = 20
+    mlflow_export: bool = False
+    mlflow_tracking_uri: str | None = None
+    mlflow_experiment: str | None = None
+    mlflow_tracing: bool = True  # Enable LLM call tracing by default
 
 
 class CompareModelsResponse(BaseModel):
@@ -454,20 +465,73 @@ async def run_scenario(request: RunScenarioRequest) -> RunScenarioResponse:
         spec = load_unified_scenario(scenario_path)
         runner = UnifiedRunner()
 
-        result = await runner.run(
-            scenario=spec,
-            model=request.model,
-            variables=request.variables,
-            max_turns=request.max_turns,
-            max_tokens=request.max_tokens,
-            temperature=request.temperature,
-        )
+        # Setup MLflow if requested
+        mlflow_config = None
+        if request.mlflow_export:
+            try:
+                from sandboxy.mlflow import MLflowConfig
+
+                mlflow_config = MLflowConfig.resolve(
+                    cli_export=True,
+                    cli_tracking_uri=request.mlflow_tracking_uri,
+                    cli_experiment=request.mlflow_experiment,
+                    cli_tracing=request.mlflow_tracing,
+                    yaml_config=None,
+                    scenario_name=spec.name,
+                )
+            except ImportError:
+                pass  # MLflow not installed
+
+        # Run with MLflow context if enabled (connects traces to run)
+        if mlflow_config and mlflow_config.enabled:
+            from sandboxy.mlflow import MLflowExporter, mlflow_run_context
+            from sandboxy.mlflow.tracing import enable_tracing
+
+            if mlflow_config.tracing:
+                enable_tracing(
+                    tracking_uri=mlflow_config.tracking_uri,
+                    experiment_name=mlflow_config.experiment,
+                )
+
+            with mlflow_run_context(mlflow_config, run_name=request.model) as run_id:
+                result = await runner.run(
+                    scenario=spec,
+                    model=request.model,
+                    variables=request.variables,
+                    max_turns=request.max_turns,
+                    max_tokens=request.max_tokens,
+                    temperature=request.temperature,
+                )
+
+                if run_id:
+                    exporter = MLflowExporter(mlflow_config)
+                    exporter.log_to_active_run(
+                        result=result,
+                        scenario_path=scenario_path,
+                        scenario_name=spec.name,
+                        scenario_id=spec.id,
+                        agent_name=request.model,
+                    )
+        else:
+            result = await runner.run(
+                scenario=spec,
+                model=request.model,
+                variables=request.variables,
+                max_turns=request.max_turns,
+                max_tokens=request.max_tokens,
+                temperature=request.temperature,
+            )
 
         # Save result to runs/
         from sandboxy.local.results import save_run_result
 
         save_run_result(request.scenario_id, result.to_dict())
 
+        # Calculate cost
+        input_tokens = result.input_tokens or 0
+        output_tokens = result.output_tokens or 0
+        cost_usd = calculate_cost(result.model, input_tokens, output_tokens)
+
         return RunScenarioResponse(
             id=result.id,
             scenario_id=result.scenario_id,
@@ -481,6 +545,9 @@ async def run_scenario(request: RunScenarioRequest) -> RunScenarioResponse:
             final_state=result.final_state,
             evaluation=result.evaluation.to_dict() if result.evaluation else None,
             latency_ms=result.latency_ms,
+            input_tokens=input_tokens,
+            output_tokens=output_tokens,
+            cost_usd=cost_usd,
             error=result.error,
         )
 
@@ -530,6 +597,19 @@ async def compare_models(request: CompareModelsRequest) -> CompareModelsResponse
 
         spec = load_unified_scenario(scenario_path)
 
+        # Enable MLflow tracing if requested (must be done BEFORE any LLM calls)
+        if request.mlflow_export and request.mlflow_tracing:
+            try:
+                from sandboxy.mlflow.tracing import enable_tracing
+
+                experiment = request.mlflow_experiment or spec.name
+                enable_tracing(
+                    tracking_uri=request.mlflow_tracking_uri,
+                    experiment_name=experiment,
+                )
+            except ImportError:
+                pass  # MLflow not installed
+
         comparison = await run_comparison(
             scenario=spec,
             models=request.models,
@@ -538,6 +618,31 @@ async def compare_models(request: CompareModelsRequest) -> CompareModelsResponse
             max_turns=request.max_turns,
         )
 
+        # MLflow export (if enabled)
+        if request.mlflow_export:
+            try:
+                from sandboxy.mlflow import MLflowConfig, MLflowExporter
+
+                for result in comparison.results:
+                    config = MLflowConfig.resolve(
+                        cli_export=True,
+                        cli_tracking_uri=request.mlflow_tracking_uri,
+                        cli_experiment=request.mlflow_experiment,
+                        cli_tracing=request.mlflow_tracing,
+                        yaml_config=None,
+                        scenario_name=spec.name,
+                    )
+                    exporter = MLflowExporter(config)
+                    exporter.export(
+                        result=result.to_dict(),
+                        scenario_path=scenario_path,
+                        agent_name=result.model,
+                    )
+            except ImportError:
+                logger.warning("MLflow not installed, skipping export")
+            except Exception as e:
+                logger.warning(f"Failed to export to MLflow: {e}")
+
         # Save comparison result
         from sandboxy.local.results import save_run_result
 
@@ -905,6 +1010,8 @@ class RunDatasetRequest(BaseModel):
     max_tokens: int = 1024
     temperature: float = 0.7
     parallel: int = 1
+    mlflow_enabled: bool = False
+    mlflow_experiment: str | None = None
 
 
 class RunDatasetResponse(BaseModel):
@@ -1335,25 +1442,81 @@ async def run_with_dataset(request: RunDatasetRequest) -> RunDatasetResponse:
         spec = load_unified_scenario(scenario_path)
         dataset = load_dataset(dataset_path)
 
-        if request.parallel > 1:
-            result = await run_dataset_parallel(
+        # Setup MLflow if enabled
+        mlflow_config = None
+        if request.mlflow_enabled:
+            try:
+                from sandboxy.mlflow import MLflowConfig
+
+                mlflow_config = MLflowConfig(
+                    enabled=True,
+                    experiment=request.mlflow_experiment or f"{spec.name}-dataset",
+                    tracing=False,  # Tracing not needed for dataset aggregates
+                )
+            except ImportError:
+                pass  # MLflow not installed
+
+        async def run_dataset_benchmark():
+            if request.parallel > 1:
+                return await run_dataset_parallel(
+                    scenario=spec,
+                    model=request.model,
+                    dataset=dataset,
+                    max_turns=request.max_turns,
+                    max_tokens=request.max_tokens,
+                    temperature=request.temperature,
+                    max_concurrent=request.parallel,
+                )
+            return await run_dataset(
                 scenario=spec,
                 model=request.model,
                 dataset=dataset,
                 max_turns=request.max_turns,
                 max_tokens=request.max_tokens,
                 temperature=request.temperature,
-                max_concurrent=request.parallel,
             )
+
+        # Run with MLflow context if enabled
+        if mlflow_config and mlflow_config.enabled:
+            from sandboxy.mlflow import mlflow_run_context
+
+            run_name = f"{request.model}-{request.dataset_id}"
+            with mlflow_run_context(mlflow_config, run_name=run_name) as run_id:
+                result = await run_dataset_benchmark()
+
+                # Log aggregate metrics to MLflow
+                if run_id:
+                    try:
+                        import mlflow
+
+                        mlflow.log_params(
+                            {
+                                "scenario_id": result.scenario_id,
+                                "dataset_id": result.dataset_id,
+                                "model": result.model,
+                                "total_cases": result.total_cases,
+                            }
+                        )
+                        mlflow.log_metrics(
+                            {
+                                "passed_cases": result.passed_cases,
+                                "failed_cases": result.failed_cases,
+                                "pass_rate": result.pass_rate,
+                                "avg_score": result.avg_score,
+                                "avg_percentage": result.avg_percentage,
+                                "total_time_ms": result.total_time_ms,
+                            }
+                        )
+                        # Log per-expected-outcome metrics
+                        for expected, counts in result.by_expected.items():
+                            total = counts.get("passed", 0) + counts.get("failed", 0)
+                            if total > 0:
+                                rate = counts.get("passed", 0) / total
+                                mlflow.log_metric(f"pass_rate_{expected}", rate)
+                    except Exception as e:
+                        logger.warning(f"Failed to log dataset metrics to MLflow: {e}")
         else:
-            result = await run_dataset(
-                scenario=spec,
-                model=request.model,
-                dataset=dataset,
-                max_turns=request.max_turns,
-                max_tokens=request.max_tokens,
-                temperature=request.temperature,
-            )
+            result = await run_dataset_benchmark()
 
         # Save result
         from sandboxy.local.results import save_run_result