feat: Add .env.example configuration file and integrate Redis service into Docker Compose for improved task management. Update desktop application to support document indexing and extraction workflows, enhancing user experience with a new extraction wizard and stepper components. Revise README and documentation for clarity on setup and usage.

services/extraction/domain/estimator.py

@@ -110,6 +110,60 @@ class Estimator:
     def _accumulate_usage(self, agg, new_usage):
         accumulate_usage(agg, new_usage)
 
+    def estimate_from_file(
+        self,
+        file_path: str,
+        doc_count: int = 1,
+        selected_paths: Optional[List[str]] = None,
+        tiff_page_range: Optional[tuple] = None,
+        model: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Estima tokens y costos procesando un único archivo subido desde el cliente.
+
+        Si `model` se especifica y es distinto al modelo actual del extractor,
+        crea un extractor temporal con ese modelo para obtener el uso real de tokens.
+        """
+        from api.desktop_ingest import _service_for_model
+
+        service = _service_for_model(self.service, model)
+        prof = get_profile('invoice')
+        paths = selected_paths or []
+        prompt = PromptBuilder.build(paths, {}, profile_name=prof.get('name', 'invoice'))
+
+        data, usage_ext = service.extract_full_from_pdf(
+            file_path,
+            instruction=prompt,
+            tiff_page_range=tiff_page_range,
+        )
+        if isinstance(data, dict):
+            data['radicado'] = 'sample'
+        else:
+            data = {"radicado": 'sample', "data": data}
+
+        norm = normalize_to_schema(data, paths=paths, profile_name=prof.get('name', 'invoice'))
+        usage_norm = None
+        if isinstance(norm, dict):
+            usage_norm = norm.get("_usage")
+
+        result = self._build_result(usage_ext, usage_norm, doc_count, 'sample')
+
+        # Enriquecer con modelo efectivo y conteo de páginas para el frontend
+        effective_model = model or getattr(service.extractor, 'model', None) or getattr(self.service.extractor, 'model', None)
+        result['extractor_model'] = effective_model
+        result['normalizer_model'] = settings.api.normalizer_model or 'gpt-5.4-mini'
+
+        # Estimar numero de paginas a partir de prompt_tokens de imagen:
+        # GPT-4o cobra ~85 tok base + ~170 tok por tile 512x512.
+        # Una A4 a 150 DPI (~1240x1754) genera ~6 tiles → ~1105 tokens.
+        # Aproximamos: pages = max(1, round(image_tokens / 1000))
+        prompt_tokens = int((usage_ext or {}).get('prompt_tokens') or 0)
+        text_prompt_est = len(prompt) // 4  # 1 token ≈ 4 chars
+        image_tokens_est = max(0, prompt_tokens - text_prompt_est)
+        pages_est = max(1, round(image_tokens_est / 1000))
+        result['pages_per_doc'] = pages_est
+
+        return result
+
     def _build_result(self, usage_ext, usage_norm, total_docs, first_radicado):
         model_ext = getattr(self.service.extractor, "model", None)
         cost_ext = compute_cost_from_usage(usage_ext, "extract", model=model_ext)