iatoolkit 1.9.0__py3-none-any.whl → 1.15.3__py3-none-any.whl

iatoolkit/infra/llm_providers/openai_adapter.py

@@ -7,8 +7,10 @@ import logging
 from typing import Dict, List, Optional
 from iatoolkit.infra.llm_response import LLMResponse, ToolCall, Usage
 from iatoolkit.common.exceptions import IAToolkitException
-import html
 from typing import List
+import mimetypes
+import re
+
 
 class OpenAIAdapter:
     """Adaptador para la API de OpenAI"""
@@ -24,9 +26,14 @@ class OpenAIAdapter:
                         tools: Optional[List[Dict]] = None,
                         text: Optional[Dict] = None,
                         reasoning: Optional[Dict] = None,
-                        tool_choice: str = "auto") -> LLMResponse:
+                        tool_choice: str = "auto",
+                        images: Optional[List[Dict]] = None) -> LLMResponse:
         """Llamada a la API de OpenAI y mapeo a estructura común"""
         try:
+            # Handle multimodal input if images are present
+            if images:
+                input = self._prepare_multimodal_input(input, images)
+
             # Preparar parámetros para OpenAI
             params = {
                 'model': model,
@@ -56,28 +63,177 @@ class OpenAIAdapter:
 
             raise IAToolkitException(IAToolkitException.ErrorType.LLM_ERROR, error_message)
 
+    def _prepare_multimodal_input(self, messages: List[Dict], images: List[Dict]) -> List[Dict]:
+        """
+        Transforma el mensaje del usuario de texto simple a contenido multimodal (texto + imágenes)
+        usando el formato de Responses API (input_text/input_image).
+        """
+        # Encontrar el último mensaje del usuario
+        target_message = None
+        for msg in reversed(messages):
+            if msg.get('role') == 'user':
+                target_message = msg
+                break
+
+        if not target_message:
+            return messages
+
+        text_content = target_message.get('content', '')
+        content_parts = []
+
+        # Agregar parte de texto (Responses API)
+        if text_content:
+            content_parts.append({"type": "input_text", "text": text_content})
+
+        # Agregar partes de imagen (Responses API)
+        for img in images:
+            filename = img.get('name', '')
+            mime_type, _ = mimetypes.guess_type(filename)
+            if not mime_type:
+                mime_type = 'image/jpeg'
+
+            base64_data = img.get('base64', '')
+            url = f"data:{mime_type};base64,{base64_data}"
+
+            content_parts.append({
+                "type": "input_image",
+                "image_url": url
+            })
+
+        # Construir nueva lista de mensajes con el contenido actualizado
+        final_messages = []
+        for msg in messages:
+            if msg is target_message:
+                new_msg = msg.copy()
+                new_msg['content'] = content_parts
+                final_messages.append(new_msg)
+            else:
+                final_messages.append(msg)
+
+        return final_messages
+
     def _map_openai_response(self, openai_response) -> LLMResponse:
         """Mapear respuesta de OpenAI a estructura común"""
-        # Mapear tool calls
-        tool_calls = []
-        if hasattr(openai_response, 'output') and openai_response.output:
-            for tool_call in openai_response.output:
-                if hasattr(tool_call, 'type') and tool_call.type == "function_call":
-                    tool_calls.append(ToolCall(
-                        call_id=getattr(tool_call, 'call_id', ''),
-                        type=tool_call.type,
-                        name=getattr(tool_call, 'name', ''),
-                        arguments=getattr(tool_call, 'arguments', '{}')
-                    ))
-
-        # Mapear usage
+        tool_calls: List[ToolCall] = []
+        content_parts: List[Dict] = []
+        output_text = ""
+
+        print(f'openai_response.output: {openai_response.output}')
+        output_items = getattr(openai_response, 'output', []) or []
+
+        def _extract_markdown_images(text: str) -> None:
+            # Pattern: ![Alt](https://...)
+            markdown_images = re.findall(r'!\[([^\]]*)\]\((https?://[^)]+)\)', text or "")
+            for _alt_text, url in markdown_images:
+                content_parts.append({
+                    "type": "image",
+                    "source": {
+                        "type": "url",
+                        "media_type": "image/webp",
+                        "url": url
+                    }
+                })
+
+        for item in output_items:
+            item_type = getattr(item, 'type', '')
+
+            # 1) Tool calls (Responses API)
+            if item_type == "function_call":
+                tool_calls.append(ToolCall(
+                    call_id=getattr(item, 'call_id', ''),
+                    type=item_type,
+                    name=getattr(item, 'name', ''),
+                    arguments=getattr(item, 'arguments', '{}')
+                ))
+                continue
+
+            # 2) Mensajes (lo más común en Responses API)
+            if item_type == "message":
+                msg_content = getattr(item, "content", None) or []
+                for part in msg_content:
+                    part_type = getattr(part, "type", "") or ""
+
+                    # 2.A) Texto
+                    if part_type in ("output_text", "text"):
+                        text_content = getattr(part, "text", "") or ""
+                        if text_content:
+                            _extract_markdown_images(text_content)
+                            output_text += text_content
+                            content_parts.append({"type": "text", "text": text_content})
+
+                    # 2.B) Imagen (puede venir como URL o base64 según el SDK/endpoint)
+                    elif part_type in ("output_image", "image"):
+                        # Algunas variantes comunes:
+                        # - part.image_url (string URL)
+                        # - part.url
+                        # - part.b64_json (base64)
+                        image_url = getattr(part, "image_url", None) or getattr(part, "url", None)
+                        b64 = getattr(part, "b64_json", None) or getattr(part, "image", None) or getattr(part, "data", None)
+
+                        # mime_type a veces viene, a veces no
+                        mime_type = getattr(part, "media_type", None) or getattr(part, "mime_type", None) or "image/png"
+
+                        if image_url:
+                            content_parts.append({
+                                "type": "image",
+                                "source": {
+                                    "type": "url",
+                                    "media_type": mime_type,
+                                    "url": image_url
+                                }
+                            })
+                            output_text += "\n[Imagen Generada]\n"
+                        elif b64:
+                            content_parts.append({
+                                "type": "image",
+                                "source": {
+                                    "type": "base64",
+                                    "media_type": mime_type,
+                                    "data": b64
+                                }
+                            })
+                            output_text += "\n[Imagen Generada]\n"
+
+                continue
+
+            # 3) Compatibilidad hacia atrás: ítems planos "text"
+            if item_type == "text":
+                text_content = getattr(item, 'text', '') or ""
+                if text_content:
+                    _extract_markdown_images(text_content)
+                    output_text += text_content
+                    content_parts.append({"type": "text", "text": text_content})
+                continue
+
+            # 4) Compatibilidad hacia atrás: ítems planos "image"
+            if item_type == "image":
+                base64_data = getattr(item, 'image', '') or getattr(item, 'data', '')
+                mime_type = getattr(item, 'media_type', 'image/png')
+                if base64_data:
+                    content_parts.append({
+                        "type": "image",
+                        "source": {
+                            "type": "base64",
+                            "media_type": mime_type,
+                            "data": base64_data
+                        }
+                    })
+                    output_text += "\n[Imagen Generada]\n"
+                continue
+
+        # Fallback: Si no se extrajo texto, probamos output_text directo
+        if not output_text:
+            output_text = getattr(openai_response, 'output_text', '') or ""
+            if output_text and not content_parts:
+                _extract_markdown_images(output_text)
+                content_parts.append({"type": "text", "text": output_text})
+
         usage = Usage(
             input_tokens=openai_response.usage.input_tokens if openai_response.usage else 0,
             output_tokens=openai_response.usage.output_tokens if openai_response.usage else 0,
             total_tokens=openai_response.usage.total_tokens if openai_response.usage else 0
         )
 
-        # Reasoning content extracted from Responses output items (type="reasoning")
         reasoning_list = self._extract_reasoning_content(openai_response)
         reasoning_str = "\n".join(reasoning_list)
 
@@ -85,10 +241,11 @@ class OpenAIAdapter:
             id=openai_response.id,
             model=openai_response.model,
             status=openai_response.status,
-            output_text=getattr(openai_response, 'output_text', ''),
+            output_text=output_text,
             output=tool_calls,
             usage=usage,
-            reasoning_content=reasoning_str
+            reasoning_content=reasoning_str,
+            content_parts=content_parts
         )
 
     def _extract_reasoning_content(self, openai_response) -> List[str]:

iatoolkit/infra/llm_response.py

@@ -34,6 +34,9 @@ class LLMResponse:
     usage: Usage
     reasoning_content: str = None # campo opcional para Chain of Thought
 
+    # ordered list of content blocks (text and image mixed)
+    # Example: [{"type": "text", "text": "..."}, {"type": "image", "source": {"type": "base64", "data": "..."}}]
+    content_parts: List[Dict] = None
 
     def __post_init__(self):
         """Asegura que output sea una lista"""
@@ -43,3 +46,13 @@ class LLMResponse:
         if self.reasoning_content is None:
             self.reasoning_content = ""
 
+        if self.content_parts is None:
+            self.content_parts = []
+
+            # if the response has legacy text and no content parts, create a default text part
+            if self.output_text:
+                self.content_parts.append({
+                    "type": "text",
+                    "text": self.output_text
+                })
+

iatoolkit/locales/en.yaml

@@ -1,5 +1,14 @@
 # Language: English
 ui:
+  common:
+    actions: "Actions"
+    save: "Save"
+    cancel: "Cancel"
+    apply: "Apply Changes"
+    discard_changes: "Discard Changes"
+    order: "Order"
+
+
   login_widget:
     title: "Sign In"
     welcome_message: "Enter your credentials or register to access this platform."
@@ -88,7 +97,15 @@ ui:
     add: "Add"
     create: "Create"
     delete: "Delete"
-
+    add_new: "Add new"
+    category_name_placeholder: "Category name"
+    items: "Items"
+    drag_drop_hint: "Use arrows to reorder items."
+    current_items: "Current items"
+    manage: "Manage"
+    unsaved_changes_title: "Unsaved Changes"
+    unsaved_changes_message: "You have unsaved changes. Do you want to save them before proceeding?"
+    discard_changes: "Discard Changes"
 
   db_explorer:
     data_explorer: "Data Explorer"
@@ -181,10 +198,32 @@ ui:
     delete_cancel: "Cancel"
     target_collection: "Collection"
     select_collection_placeholder: "Select a collection"
+    select_collection: "Select a collection"
     collection_required: "Collection is required"
     collection: "Collection"
+    manage_collections: "Manage collections"
     all_collections: "All collections"
 
+  json_editor:
+    title: "JSON Schema Editor"
+    subtitle: "Define the structure and metadata for this JSON field"
+    field_name: "Field Name"
+    type: "Data Type"
+    description: "Description"
+    empty_schema: "No fields defined yet."
+    add_root: "Add Root Field"
+    add_field: "Add Field"
+    types:
+      string: "String"
+      integer: "Integer"
+      number: "Number"
+      boolean: "Boolean"
+      object: "Object (Map)"
+      array: "Array (List)"
+      jsonb: "JSON Raw"
+
+
+
   tooltips:
     history: "History of my queries"
     reload_context: "Force Context Reload"
@@ -350,7 +389,6 @@ js_messages:
   db_created: "Created"
   db_last_access: "Last access"
   db_filename: "Filename"
-  db_user: "User"
   db_status: "Status"
   db_collection: "Collection"
   editor_no_file_selected: "No file selected"
@@ -358,6 +396,13 @@ js_messages:
   cant_load_company: "Could not load company.yaml"
   config_saved: "Configuration saved successfully."
   config_error: "Error saving configuration."
+  all_collections: "All collections"
+  no_variables_found: "No variables found in prompt template."
+  manage_collections: "Manage collections"
+  edit: "Edit"
+
+
+
 
 
 

iatoolkit/locales/es.yaml

@@ -1,5 +1,13 @@
 # locales/es.yaml
 ui:
+  common:
+    actions: "Acciones"
+    save: "Guardar"
+    cancel: "Cancelar"
+    apply: "Aplicar Cambios"
+    discard_changes: "Descartar Cambios"
+    order: "Orden"
+
   login_widget:
     title: "Iniciar Sesión"
     welcome_message: "Ingresa tus credenciales o registrate para acceder a la plataforma."
@@ -85,6 +93,16 @@ ui:
     add: "Agregar"
     create: "Crear"
     delete: "Borrar"
+    add_new: "Agregar nuevo"
+    category_name_placeholder: "nombre de categoria"
+    items: "Items"
+    drag_drop_hint: "Utiliza las flechas para ordenar items.."
+    current_items: "items actuales"
+    manage: "Administrar"
+    unsaved_changes_title: "Cambios sin guardar"
+    unsaved_changes_message: "Tienes modificaciones sin guardar. Quieres grabarlos antes de salir?"
+    discard_changes: "Descartar cambios"
+
 
   db_explorer:
     data_explorer: "Explorador de datos"
@@ -136,6 +154,7 @@ ui:
     category_label: "Categoría"
     delete_confirmation: "Eliminar el prompt?"
 
+
   config:
     editor_description: "Editor de configuración"
     title: "Editor de configuraciones"
@@ -179,7 +198,26 @@ ui:
     collection_required: "Debe seleccionar una categoría"
     all_collections: "Todas las categorías"
     collection: "Categoría"
+    manage_collections: "Administra collections"
+    select_collection: "Selecciona una categorìa"
 
+  json_editor:
+    title: "Editor de Esquema JSON"
+    subtitle: "Define la estructura y metadatos para este campo JSON"
+    field_name: "Nombre del Campo"
+    type: "Tipo de Dato"
+    description: "Descripción"
+    empty_schema: "Aún no hay campos definidos."
+    add_root: "Agregar Campo Raíz"
+    add_field: "Agregar Campo"
+    types:
+      string: "Texto (String)"
+      integer: "Entero"
+      number: "Número (Decimal)"
+      boolean: "Booleano"
+      object: "Objeto (Mapa)"
+      array: "Lista (Array)"
+      jsonb: "JSON Crudo"
 
   tooltips:
     history: "Historial con mis consultas"
@@ -340,7 +378,6 @@ js_messages:
   search_placeholder: "Buscar usuarios..."
   showing: "Mostrando"
   records: "Registros"
-  db_user: "Usuario"
   db_role: "Rol"
   db_verified: "Verificado"
   db_collection: "Colección"
@@ -354,6 +391,13 @@ js_messages:
   cant_load_company: "No puede cargarcompany.yaml"
   config_saved: "Configuración guardada correctamente."
   config_error: "Error guardando configuración"
+  all_collections: "Todas las colecciones"
+  no_variables_found: "El template no contiene variables."
+  manage_collections: "Administra collections"
+  edit: "Editar"
+
+
+
 
 
 

iatoolkit/repositories/llm_query_repo.py

@@ -22,12 +22,20 @@ class LLMQueryRepo:
     def rollback(self):
         self.session.rollback()
 
+    # save new query result in the database
     def add_query(self, query: LLMQuery):
         self.session.add(query)
         self.session.commit()
         return query
 
+    # get user query history
+    def get_history(self, company: Company, user_identifier: str) -> list[LLMQuery]:
+        return self.session.query(LLMQuery).filter(
+            LLMQuery.user_identifier == user_identifier,
+        ).filter_by(company_id=company.id).order_by(LLMQuery.created_at.desc()).limit(100).all()
 
+
+    ## --- Tools related methods
     def get_company_tools(self, company: Company) -> list[Tool]:
         return (
             self.session.query(Tool)
@@ -63,6 +71,30 @@ class LLMQueryRepo:
     def delete_tool(self, tool: Tool):
         self.session.query(Tool).filter_by(id=tool.id).delete(synchronize_session=False)
 
+    # -- Prompt related methods
+
+    def get_prompt_by_name(self, company: Company, prompt_name: str):
+        return self.session.query(Prompt).filter_by(company_id=company.id, name=prompt_name).first()
+
+
+    def get_prompts(self, company: Company, include_all: bool = False) -> list[Prompt]:
+        if include_all:
+            # Include all prompts (for the prompt admin dashboard)
+            return self.session.query(Prompt).filter(
+                Prompt.company_id == company.id,
+            ).all()
+        else:
+            # Only active company prompts (default behavior for end users)
+            return self.session.query(Prompt).filter(
+                Prompt.company_id == company.id,
+                Prompt.prompt_type == PromptType.COMPANY.value,
+                Prompt.active == True
+            ).all()
+
+    def get_system_prompts(self) -> list[Prompt]:
+        return self.session.query(Prompt).filter_by(prompt_type=PromptType.SYSTEM.value, active=True).order_by(
+            Prompt.order).all()
+
     def create_or_update_prompt(self, new_prompt: Prompt):
         prompt = self.session.query(Prompt).filter_by(company_id=new_prompt.company_id,
                                                  name=new_prompt.name).first()
@@ -80,6 +112,18 @@ class LLMQueryRepo:
         self.session.commit()
         return prompt
 
+    def delete_prompt(self, prompt: Prompt):
+        self.session.delete(prompt)
+        self.session.commit()
+
+    # -- Prompt category methods
+
+    def get_category_by_name(self, company_id: int, name: str) -> PromptCategory:
+        return self.session.query(PromptCategory).filter_by(company_id=company_id, name=name).first()
+
+    def get_all_categories(self, company_id: int) -> List[PromptCategory]:
+        return self.session.query(PromptCategory).filter_by(company_id=company_id).order_by(PromptCategory.order).all()
+
     def create_or_update_prompt_category(self, new_category: PromptCategory):
         category = self.session.query(PromptCategory).filter_by(company_id=new_category.company_id,
                                                       name=new_category.name).first()
@@ -92,36 +136,3 @@ class LLMQueryRepo:
         self.session.flush()
         return category
 
-    def get_history(self, company: Company, user_identifier: str) -> list[LLMQuery]:
-        return self.session.query(LLMQuery).filter(
-            LLMQuery.user_identifier == user_identifier,
-        ).filter_by(company_id=company.id).order_by(LLMQuery.created_at.desc()).limit(100).all()
-
-    def get_prompts(self, company: Company, include_all: bool = False) -> list[Prompt]:
-        if include_all:
-            # Include all  prompts: company, system, agent
-            return self.session.query(Prompt).filter(
-                Prompt.company_id == company.id,
-            ).all()
-        else:
-            # Only company prompts, excluding system (default behavior for end users)
-            return self.session.query(Prompt).filter(
-                Prompt.company_id == company.id,
-                Prompt.prompt_type == PromptType.COMPANY.value
-            ).all()
-
-    def get_prompt_by_name(self, company: Company, prompt_name: str):
-        return self.session.query(Prompt).filter_by(company_id=company.id, name=prompt_name).first()
-
-    def get_category_by_name(self, company_id: int, name: str) -> PromptCategory:
-        return self.session.query(PromptCategory).filter_by(company_id=company_id, name=name).first()
-
-    def get_all_categories(self, company_id: int) -> List[PromptCategory]:
-        return self.session.query(PromptCategory).filter_by(company_id=company_id).order_by(PromptCategory.order).all()
-
-    def get_system_prompts(self) -> list[Prompt]:
-        return self.session.query(Prompt).filter_by(prompt_type=PromptType.SYSTEM.value, active=True).order_by(Prompt.order).all()
-
-    def delete_prompt(self, prompt: Prompt):
-        self.session.delete(prompt)
-        self.session.commit()