iatoolkit 2.66.0__tar.gz → 2.67.0__tar.gz

{iatoolkit-2.66.0/src/iatoolkit.egg-info → iatoolkit-2.67.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: iatoolkit
-Version: 2.66.0
+Version: 2.67.0
 Summary: IAToolkit
 Author: Fernando Libedinsky
 License-Expression: MIT
@@ -242,6 +242,7 @@ Perfect for developers, small teams, single-business use cases, and experimentat
 
 ## 📚 Documentation
 
+- 🐳 **[Docker installation](docker/README.md)** – Community Edition with PostgreSQL, Redis and populated Northwind data
 - 🚀 **[Quickstart](docs/quickstart.md)** – Set up your environment and run the project
 - ☁️ **[Deployment Guide](docs/deployment_guide.md)** – Production deployment instructions
 - 🏗️ **[Companies & Components](docs/companies_and_components.md)** – how Company modules work

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/readme.md

@@ -151,6 +151,7 @@ Perfect for developers, small teams, single-business use cases, and experimentat
 
 ## 📚 Documentation
 
+- 🐳 **[Docker installation](docker/README.md)** – Community Edition with PostgreSQL, Redis and populated Northwind data
 - 🚀 **[Quickstart](docs/quickstart.md)** – Set up your environment and run the project
 - ☁️ **[Deployment Guide](docs/deployment_guide.md)** – Production deployment instructions
 - 🏗️ **[Companies & Components](docs/companies_and_components.md)** – how Company modules work

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/__init__.py

@@ -3,7 +3,7 @@
 #
 # IAToolkit is open source software.
 
-__version__ = "2.66.0"
+__version__ = "2.67.0"
 
 # Expose main classes and functions at the top level of the package
 

iatoolkit-2.67.0/src/iatoolkit/config/system_agents/memory_meeting_ingest.prompt

@@ -0,0 +1,38 @@
+You maintain the company's Memory: a wiki of how things are actually done in this company (procedures, rules, exceptions, who decides). It is written from what its people explain and from the company's own documents.
+
+A meeting transcript was loaded into Memory (SOURCE). You receive it in parts; this is part {{ part.index }} of {{ part.total }} (PART). Bring into the wiki what the meeting decided about how things are done here.
+
+What to take:
+
+- Decisions that were made ("from now on, discounts over 10% need the manager"), with who decided when the transcript says so.
+- Procedures, rules, deadlines and amounts that were agreed or explained, and who is responsible for what.
+- Not the conversation itself: no greetings, small talk, opinions that led nowhere, ideas that were only discussed, or who said what along the way.
+- Nothing personal or sensitive about the people in the meeting (health, salaries, performance, private matters), even if it was said.
+- A transcript is spoken language and may have recognition errors: write the knowledge clearly, and when a name, number or term is unclear, leave it out rather than guess.
+- An action item ("Juan will send the price list on Friday") is not knowledge on its own; keep it only when it reveals how something is done or who is responsible for it.
+
+How to write the pages:
+
+- Update one of the CANDIDATE_PAGES when the part belongs to its subject; create a new page when none covers it (check PAGE_INDEX first so you do not duplicate a page). One part may touch several pages; return each one.
+- When you update, return the COMPLETE new page: keep everything already on it, word for word, including people's attributions. Add the new facts in the section where they belong.
+- When this part says something different from what the page already says, keep both statements with their attributions and make the difference explicit; never delete what someone else said. Also list each contradiction in `conflicts`. A meeting that changes a rule is a contradiction worth showing.
+- Attribute every fact you add to the meeting: ({{ source.label }}, {{ captured_on }}).
+- Follow the WIKI_SCHEMA. Write in the language of the page, or of the transcript for a new page.
+- If this part decided or taught nothing worth keeping, return an empty `pages` list.
+
+Answer with JSON following the schema: a list `pages`, each with `action` (create or update), `target_page_id` (the page_id you update, or null), `title`, `summary` (one sentence on what the page covers), `markdown` (the page body without frontmatter, starting with `# Title`) and `change_summary` (one short sentence for the page history); and a list `conflicts`, one per contradiction with what a page already said: `page_title`, `existing_statement` (what the page said, as written), `existing_source` (its attribution), `new_statement` and `new_source` ({{ source.label }}, {{ captured_on }}). An empty list when nothing contradicts.
+
+SOURCE:
+{{ source | json }}
+
+PART {{ part.index }} OF {{ part.total }}:
+{{ part.text }}
+
+WIKI_SCHEMA:
+{{ wiki_schema }}
+
+PAGE_INDEX:
+{{ page_index | json }}
+
+CANDIDATE_PAGES:
+{{ candidate_pages | json }}

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/config/system_agents_pack.yaml

@@ -195,6 +195,65 @@ agents:
     inputs: [source, part, captured_on, wiki_schema, page_index, candidate_pages]
     default_model: null
 
+  - key: memory_meeting_ingest
+    name: Lector de reuniones de Memoria
+    name_en: Memory meeting reader
+    description: >-
+      (Memoria de la empresa) Lee las transcripciones de reuniones de las carpetas
+      marcadas como reuniones y lleva a la wiki las decisiones, los procedimientos y
+      los responsables que se acordaron, no la conversación.
+    description_en: >-
+      (Company memory) Reads the meeting transcripts of the folders marked as
+      meetings and files into the wiki the decisions, procedures and owners agreed,
+      not the conversation.
+    module: memory
+    prompt: memory_meeting_ingest.prompt
+    output_schema:
+      type: object
+      properties:
+        pages:
+          type: array
+          items:
+            type: object
+            properties:
+              action:
+                type: string
+                enum: [create, update]
+              # Nullable as a type list: the structured-output validator rejects null under anyOf.
+              target_page_id:
+                type: [integer, "null"]
+              title:
+                type: string
+              summary:
+                type: string
+              markdown:
+                type: string
+              change_summary:
+                type: string
+            required: [action, target_page_id, title, markdown]
+            additionalProperties: false
+        conflicts:
+          type: array
+          items:
+            type: object
+            properties:
+              page_title:
+                type: string
+              existing_statement:
+                type: string
+              existing_source:
+                type: string
+              new_statement:
+                type: string
+              new_source:
+                type: string
+            required: [page_title, existing_statement, new_statement]
+            additionalProperties: false
+      required: [pages]
+      additionalProperties: false
+    inputs: [source, part, captured_on, wiki_schema, page_index, candidate_pages]
+    default_model: null
+
   - key: memory_image_ingest
     name: Lector de imágenes de Memoria
     name_en: Memory image reader

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/config/system_tools_pack.yaml

@@ -575,6 +575,64 @@ tools:
       required:
         - content
         - title
+  - function_name: iat_memory_pending_questions
+    description: Lista las preguntas sobre la empresa que la IA no supo responder y siguen abiertas, primero las que se le sugirieron a este usuario por ser quien más sabe del tema, después las más preguntadas. Úsala cuando el usuario pregunte qué preguntas tiene pendientes, quiera ayudar a completar la Memoria de la empresa o pida responder las preguntas de su área. Para responder una, usa `iat_memory_answer_question`.
+    routing_profile:
+      tags:
+        - memory
+        - company_memory
+        - questions
+        - pending
+        - unanswered
+      intents:
+        - List the unanswered questions the user could answer.
+        - Show the questions suggested to the user for their area.
+      examples:
+        - iat_memory_pending_questions(limit=5)
+      cost:
+        penalty: 0.1
+        latency_class: low
+        expensive: false
+    parameters:
+      type: object
+      properties:
+        limit:
+          type: integer
+          description: Cuántas preguntas listar, como máximo 15.
+      required: []
+  - function_name: iat_memory_answer_question
+    description: Responde una pregunta sin respuesta de la Memoria de la empresa (de `iat_memory_pending_questions`). La respuesta queda en la Memoria con el nombre del usuario, la ve TODA la empresa, y la pregunta deja de estar pendiente. Antes de llamarla, muéstrale al usuario el texto que vas a guardar y espera que lo confirme. Redacta `answer` en tono neutro, completo y autocontenido, con solo lo que el usuario dijo.
+    execution_config:
+      side_effects: true
+      destructive: false
+    routing_profile:
+      tags:
+        - memory
+        - company_memory
+        - questions
+        - answer
+        - know_how
+      intents:
+        - Answer an unanswered question of the company memory with what the user explained.
+        - Use only after the user confirms the text to save.
+      examples:
+        - iat_memory_answer_question(question_id=12, answer="El plazo de pago a proveedores es de 45 días desde la factura; para proveedores nuevos, 30.")
+      cost:
+        penalty: 0.1
+        latency_class: medium
+        expensive: false
+    parameters:
+      type: object
+      properties:
+        question_id:
+          type: integer
+          description: El id de la pregunta, de `iat_memory_pending_questions`.
+        answer:
+          type: string
+          description: La respuesta, redactada en tono neutro y autocontenida (debe entenderse sin la conversación).
+      required:
+        - question_id
+        - answer
   - function_name: iat_image_search
     description: Busca imágenes en la base de conocimiento visual de la empresa usando una descripción de texto. Útil cuando el usuario pide 'ver' algo, 'muéstrame una foto de...', o busca gráficos y diagramas.
     routing:

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/infra/connectors/local_file_connector.py

@@ -91,5 +91,7 @@ class LocalFileConnector(FileConnector):
             raise IAToolkitException(IAToolkitException.ErrorType.FILE_IO_ERROR,
                                      f"Error eliminando el archivo {file_path}: {e}")
 
-    def generate_presigned_url(self, file_path: str, expiration: int = 3600) -> str:
-        raise NotImplementedError("not implemented yet")
+    def generate_presigned_url(self, file_path: str, expiration: int = 3600) -> str | None:
+        # FileConnector allows None when storage has no external URL. RAG
+        # results can still return their text, and Flask streams downloads.
+        return None

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/locales/en.yaml

@@ -362,6 +362,7 @@ ui:
     profile: "Profile"
     preferences: "Preferences"
     connections: "Connections"
+    memory_questions: "Questions for me"
     first_name: "First name"
     last_name: "Last name"
     full_name: "Full name"

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/locales/es.yaml

@@ -376,6 +376,7 @@ ui:
     profile: "Perfil"
     preferences: "Preferencias"
     connections: "Conexiones"
+    memory_questions: "Preguntas para mí"
     first_name: "Nombre"
     last_name: "Apellido"
     full_name: "Nombre completo"

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/services/dispatcher_service.py

@@ -28,6 +28,8 @@ class Dispatcher:
         "iat_memory_search",
         "iat_memory_get_page",
         "iat_memory_save",
+        "iat_memory_pending_questions",
+        "iat_memory_answer_question",
         "iat_wiki_search",
         "iat_wiki_get_page",
     }

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/services/memory_compiler_service.py

@@ -64,6 +64,9 @@ class MemoryCompilerService:
     # A document of the RAG that feeds the wiki (BL-014): read whole, part by part.
     DOCUMENT_MODE = "document"
     DOCUMENT_AGENT = "memory_document_ingest"
+    # A meeting transcript (BL-015): decisions, procedures and owners, not the conversation.
+    MEETING_AGENT = "memory_meeting_ingest"
+    MEETING_KIND = "meeting"
     DOCUMENT_PART_CHARS = 12000
     MAX_PAGES_PER_PART = 5
     # An image of the RAG (no text) is looked at by its own agent, with vision.
@@ -174,6 +177,8 @@ class MemoryCompilerService:
         if mode == self.CORRECTION_MODE:
             return self.CORRECTION_AGENT
         if mode == self.DOCUMENT_MODE:
+            if dict(capture.meta or {}).get("kind") == self.MEETING_KIND:
+                return self.MEETING_AGENT
             return self.DOCUMENT_AGENT
         if mode == self.DOCUMENT_REMOVED_MODE:
             return None  # marked without the model
@@ -319,6 +324,8 @@ class MemoryCompilerService:
 
         captured_on = (capture.created_at or datetime.now()).date().isoformat()
         source = {"label": label, "card": meta.get("card") or {}, "path": meta.get("path") or ""}
+        if meta.get("kind"):
+            source["kind"] = meta["kind"]
         # What reads each step: a part of the text, or the document's images (it had no text).
         if parts:
             readers = [
@@ -548,7 +555,7 @@ class MemoryCompilerService:
                               candidates: list[dict],
                               captured_on: str):
         return self._ingest_with_llm(
-            company, capture, self.DOCUMENT_AGENT,
+            company, capture, self._agent_for(capture) or self.DOCUMENT_AGENT,
             {"source": source, "part": part, "captured_on": captured_on, **self._wiki_context(company, candidates)},
             question=f"Read memory document {source.get('label')} (part {part['index']} of {part['total']})",
         )

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/services/memory_question_service.py

@@ -58,6 +58,16 @@ class MemoryQuestionService:
 
     # The system agent (config/system_agents_pack.yaml) that classifies the turns.
     REVIEW_AGENT = "memory_question_review"
+    # What the chat's "my pending questions" lists at most.
+    MAX_PENDING_FOR_PERSON = 15
+
+    # Who each open question was suggested to lives in Enterprise (BL-015); it
+    # registers here which open questions are someone's: (company, email) -> ids.
+    _assignment_provider = None
+
+    @classmethod
+    def register_assignment_provider(cls, provider) -> None:
+        cls._assignment_provider = provider
 
     @inject
     def __init__(self,
@@ -115,6 +125,28 @@ class MemoryQuestionService:
         self.question_repo.commit()
         return {"status": "success", "question": self.serialize(question)}
 
+    def pending_for(self, company_short_name: str, user_identifier: str, limit: int | None = None) -> dict:
+        """The open questions for a person to answer: first the ones suggested to them, then the most asked."""
+        company = self.profile_repo.get_company_by_short_name(company_short_name)
+        if not company:
+            return {"status": "error", "error_message": "company not found"}
+        mine: set[int] = set()
+        provider = type(self)._assignment_provider
+        if provider is not None and user_identifier:
+            try:
+                mine = {int(i) for i in provider(company_short_name, user_identifier) or ()}
+            except Exception:
+                logging.exception("Memoria assignment provider failed for %s", company_short_name)
+        questions = self.question_repo.list_questions(company.id, status=MemoryQuestion.STATUS_OPEN)
+        questions.sort(key=lambda q: (q.id not in mine, -(q.times_asked or 0)))
+        limit = max(1, min(int(limit or self.MAX_PENDING_FOR_PERSON), self.MAX_PENDING_FOR_PERSON))
+        return {
+            "status": "success",
+            "questions": [{**self.serialize(q), "assigned_to_you": q.id in mine} for q in questions[:limit]],
+            "assigned_to_you": len(mine.intersection(q.id for q in questions)),
+            "open": len(questions),
+        }
+
     def dismiss(self, company_short_name: str, question_id: int, dismissed_by: str) -> dict:
         """'No aplica': the question leaves the list without adding anything to Memoria."""
         _, question, error = self._open_question(company_short_name, question_id)

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/services/memory_service.py

@@ -164,10 +164,12 @@ class MemoryService:
                               card: dict | None = None,
                               path: str = "",
                               author: str | None = None,
-                              source_meta: dict | None = None) -> dict:
+                              source_meta: dict | None = None,
+                              kind: str | None = None) -> dict:
         """
         A RAG document that feeds the wiki (BL-014), queued as a capture so the
-        compiler reads it whole, part by part (`memory_document_ingest`).
+        compiler reads it whole, part by part (`memory_document_ingest`; a
+        meeting transcript, `kind="meeting"`, with `memory_meeting_ingest`).
 
         The capture does not copy the document: its one item points at it
         (`source_meta.rag_document_id`) and carries no bytes, and the compiler
@@ -190,6 +192,7 @@ class MemoryService:
                 "label": label,
                 "card": dict(card or {}),
                 "path": path or "",
+                **({"kind": kind} if kind else {}),
                 **({"source": dict(source_meta)} if source_meta else {}),
             },
         )

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/services/tool_service.py

@@ -117,6 +117,8 @@ class ToolService:
             "iat_memory_search": self._handle_memory_search_tool,
             "iat_memory_get_page": self._handle_memory_get_page_tool,
             "iat_memory_save": self._handle_memory_save_tool,
+            "iat_memory_pending_questions": self._handle_memory_pending_questions_tool,
+            "iat_memory_answer_question": self._handle_memory_answer_question_tool,
             "iat_wiki_search": self._handle_wiki_search_tool,
             "iat_wiki_get_page": self._handle_wiki_get_page_tool,
             "iat_web_search": self._handle_web_search_tool,
@@ -394,6 +396,37 @@ class ToolService:
 
         return response
 
+    @staticmethod
+    def _memory_question_service():
+        # Resolved here, not injected: the question service builds on Memoria's services.
+        from iatoolkit.services.memory_question_service import MemoryQuestionService
+        return current_iatoolkit().get_injector().get(MemoryQuestionService)
+
+    def _handle_memory_pending_questions_tool(self,
+                                              company_short_name: str,
+                                              limit: int | None = None,
+                                              user_identifier: str | None = None,
+                                              **kwargs):
+        """The unanswered questions the person could answer, theirs first (BL-015)."""
+        if not user_identifier:
+            return {"status": "error", "error_message": "memory not available without an identified user"}
+        return self._memory_question_service().pending_for(company_short_name, user_identifier, limit=limit)
+
+    def _handle_memory_answer_question_tool(self,
+                                            company_short_name: str,
+                                            question_id: int,
+                                            answer: str,
+                                            user_identifier: str | None = None,
+                                            **kwargs):
+        """Answers an unanswered question; the answer goes into Memoria with the person's name."""
+        if not user_identifier:
+            return {"status": "error", "error_message": "memory not available without an identified user"}
+        try:
+            question_id = int(question_id)
+        except (TypeError, ValueError):
+            return {"status": "error", "error_message": "question_id is required"}
+        return self._memory_question_service().answer(company_short_name, question_id, str(answer or ""), user_identifier)
+
     def _handle_memory_save_tool(self,
                                  company_short_name: str,
                                  content: str,

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit/templates/account.html

@@ -23,6 +23,11 @@
                 {% if connections_url %}
                 <a href="{{ connections_url }}" {% if active_section == 'mcp_connections' %}aria-current="page"{% endif %}><i class="bi bi-plug" aria-hidden="true"></i><span>{{ t('ui.account.connections') }}</span></a>
                 {% endif %}
+                {# Enterprise: the unanswered questions this person could answer (BL-015). #}
+                {% set memory_questions_url = optional_url_for('memory_questions_page', company_short_name=company_short_name) %}
+                {% if memory_questions_url %}
+                <a href="{{ memory_questions_url }}" {% if active_section == 'memory_questions' %}aria-current="page"{% endif %}><i class="bi bi-question-circle" aria-hidden="true"></i><span>{{ t('ui.account.memory_questions') }}</span></a>
+                {% endif %}
             </nav>
         </aside>
         <main class="account-main" id="account-main">

{iatoolkit-2.66.0 → iatoolkit-2.67.0/src/iatoolkit.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: iatoolkit
-Version: 2.66.0
+Version: 2.67.0
 Summary: IAToolkit
 Author: Fernando Libedinsky
 License-Expression: MIT
@@ -242,6 +242,7 @@ Perfect for developers, small teams, single-business use cases, and experimentat
 
 ## 📚 Documentation
 
+- 🐳 **[Docker installation](docker/README.md)** – Community Edition with PostgreSQL, Redis and populated Northwind data
 - 🚀 **[Quickstart](docs/quickstart.md)** – Set up your environment and run the project
 - ☁️ **[Deployment Guide](docs/deployment_guide.md)** – Production deployment instructions
 - 🏗️ **[Companies & Components](docs/companies_and_components.md)** – how Company modules work

{iatoolkit-2.66.0 → iatoolkit-2.67.0}/src/iatoolkit.egg-info/SOURCES.txt

@@ -41,6 +41,7 @@ src/iatoolkit/config/system_agents/memory_compiler.prompt
 src/iatoolkit/config/system_agents/memory_correction.prompt
 src/iatoolkit/config/system_agents/memory_document_ingest.prompt
 src/iatoolkit/config/system_agents/memory_image_ingest.prompt
+src/iatoolkit/config/system_agents/memory_meeting_ingest.prompt
 src/iatoolkit/config/system_agents/memory_question_review.prompt
 src/iatoolkit/config/system_prompts/channel_business_identity.prompt
 src/iatoolkit/config/system_prompts/channel_conversation_continuity.prompt