Axlfc · Axlfc · Jul 30, 2026 · Jul 30, 2026
diff --git a/README.ca.md b/README.ca.md
@@ -59,6 +59,7 @@ La plataforma està dissenyada per a desenvolupadors, científics de dades i equ
 
 - [**simplified-stack**](simplified-stack/README.ca.md): Versió lleugera optimitzada per al desenvolupament local que integra Drupal, Obsidian i Forgejo per a fluxos de treball d'IA aïllats.
 - [**very-simplified-stack**](very-simplified-stack/README.ca.md): Versió minimalista que elimina l'orquestració de n8n i es centra en serveis de veu i l'API d'agent Cognito, dissenyada per connectar amb una instància d'Ollama externa.
+- [**AGI Agents & NOOA Framework**](agents/README.md): Mòdul d'agents cognitius que implementa el Roadmap de 5 fases (Chain-of-Thought, Autovalidació i iteració, Memòria a llarg termini amb SQLite + Vectorial, Multi-agent, i Autonomia) integrat de manera nativa amb suport per al paradigma NOOA (NVIDIA-labs Object Oriented Agents).
 
 ---
 

diff --git a/README.en.md b/README.en.md
@@ -59,6 +59,7 @@ The platform is designed for developers, data scientists, and AI teams that need
 
 - [**simplified-stack**](simplified-stack/README.en.md): Lightweight version optimized for local development that integrates Drupal, Obsidian, and Forgejo for isolated AI workflows.
 - [**very-simplified-stack**](very-simplified-stack/README.en.md): Minimalist version that removes n8n orchestration and focuses on voice services and the Cognito agent API, designed to connect to an external Ollama instance.
+- [**AGI Agents & NOOA Framework**](agents/README.md): Cognitive agents module implementing the 5-phase Roadmap (Chain-of-Thought, Self-Evaluation & iteration, Long-Term SQLite + Vectorial Memory, Multi-Agent, and Autonomy) natively integrated with full support for the NOOA (NVIDIA-labs Object Oriented Agents) paradigm.
 
 ---
 

diff --git a/README.md b/README.md
@@ -58,6 +58,7 @@ La plataforma está diseñada para desarrolladores, científicos de datos y equi
 
 - [**simplified-stack**](simplified-stack/README.md): Versión ligera optimizada para desarrollo local que integra Drupal, Obsidian y Forgejo para flujos de trabajo de IA aislados.
 - [**very-simplified-stack**](very-simplified-stack/README.md): Versión minimalista que elimina la orquestación de n8n y se centra en servicios de voz y el API de agente Cognito, diseñada para conectar con una instancia de Ollama externa.
+- [**AGI Agents & NOOA Framework**](agents/README.md): Módulo de agentes cognitivos que implementa el Roadmap de 5 fases (Chain-of-Thought, Autovalidación e iteración, Memoria a largo plazo con SQLite + Vectorial, Multi-agente, y Autonomía) integrado de forma nativa con soporte para el paradigma NOOA (NVIDIA-labs Object Oriented Agents).
 
 ---
 

diff --git a/README.zh-cn.md b/README.zh-cn.md
@@ -58,7 +58,8 @@
 ## 🔄 变体版本
 
 - [**simplified-stack**](simplified-stack/README.zh-cn.md): 为本地开发优化的轻量级版本，集成了 Drupal、Obsidian 和 Forgejo，用于隔离的 AI 工作流。
-- [**very-simplified-stack**](very-simplified-stack/README.zh-cn.md): 删除了 n8n 编排的极简版本，专注于语音服务和 Cognito 代理 API，旨在连接到外部 Ollama 实例。
+- [**very-simplified-stack**](very-simplified-stack/README.zh-cn.md): 删除了 n8n 编排的极简版本，专注于语音服务 and Cognito 代理 API，旨在连接 to 外部 Ollama 实例。
+- [**AGI Agents & NOOA Framework**](agents/README.md): 认知代理模块，实现5阶段路线图（思维链、自我评估与迭代、基于SQLite + 向量的长期记忆、多智能体协作、以及完全自主），原生集成并全面支持 NOOA (NVIDIA-labs Object Oriented Agents) 框架规范。
 
 ---
 

diff --git a/agents/README.ca.md b/agents/README.ca.md
@@ -14,8 +14,8 @@ Welcome to the AGI agents module. This directory contains the implementation of
 | 1 | **Chain-of-Thought** | ✅ IMPLEMENTED | Reasoning + Multi-LLM routing |
 | 2 | **Self-Evaluation** | ✅ IMPLEMENTED | Output validation + iteration |
 | 3 | **Memory & Learning** | ✅ IMPLEMENTED | Experience storage + few-shot |
-| 4 | Multi-Agent Collab | 🔄 Planned | Team coordination |
-| 5 | Autonomous Op | 🔄 Planned | Fully autonomous loops |
+| 4 | **Multi-Agent Collab** | ✅ IMPLEMENTED | Team coordination & Orchestrator |
+| 5 | **Autonomous Op** | ✅ IMPLEMENTED | Fully autonomous loops & Optimization |
 
 ## 🚀 Quick Start
 

diff --git a/agents/README.en.md b/agents/README.en.md
@@ -14,8 +14,8 @@ Welcome to the AGI agents module. This directory contains the implementation of
 | 1 | **Chain-of-Thought** | ✅ IMPLEMENTED | Reasoning + Multi-LLM routing |
 | 2 | **Self-Evaluation** | ✅ IMPLEMENTED | Output validation + iteration |
 | 3 | **Memory & Learning** | ✅ IMPLEMENTED | Experience storage + few-shot |
-| 4 | Multi-Agent Collab | 🔄 Planned | Team coordination |
-| 5 | Autonomous Op | 🔄 Planned | Fully autonomous loops |
+| 4 | **Multi-Agent Collab** | ✅ IMPLEMENTED | Team coordination & Orchestrator |
+| 5 | **Autonomous Op** | ✅ IMPLEMENTED | Fully autonomous loops & Optimization |
 
 ## 🚀 Quick Start
 

diff --git a/agents/README.md b/agents/README.md
@@ -14,8 +14,8 @@ Welcome to the AGI agents module. This directory contains the implementation of
 | 1 | **Chain-of-Thought** | ✅ IMPLEMENTED | Reasoning + Multi-LLM routing |
 | 2 | **Self-Evaluation** | ✅ IMPLEMENTED | Output validation + iteration |
 | 3 | **Memory & Learning** | ✅ IMPLEMENTED | Experience storage + few-shot |
-| 4 | Multi-Agent Collab | 🔄 Planned | Team coordination |
-| 5 | Autonomous Op | 🔄 Planned | Fully autonomous loops |
+| 4 | **Multi-Agent Collab** | ✅ IMPLEMENTED | Team coordination & Orchestrator |
+| 5 | **Autonomous Op** | ✅ IMPLEMENTED | Fully autonomous loops & Optimization |
 
 ## 🚀 Quick Start
 

diff --git a/agents/README.zh-cn.md b/agents/README.zh-cn.md
@@ -14,8 +14,8 @@ Welcome to the AGI agents module. This directory contains the implementation of
 | 1 | **Chain-of-Thought** | ✅ IMPLEMENTED | Reasoning + Multi-LLM routing |
 | 2 | **Self-Evaluation** | ✅ IMPLEMENTED | Output validation + iteration |
 | 3 | **Memory & Learning** | ✅ IMPLEMENTED | Experience storage + few-shot |
-| 4 | Multi-Agent Collab | 🔄 Planned | Team coordination |
-| 5 | Autonomous Op | 🔄 Planned | Fully autonomous loops |
+| 4 | **Multi-Agent Collab** | ✅ IMPLEMENTED | Team coordination & Orchestrator |
+| 5 | **Autonomous Op** | ✅ IMPLEMENTED | Fully autonomous loops & Optimization |
 
 ## 🚀 Quick Start
 

diff --git a/agents/agent_router.py b/agents/agent_router.py
@@ -30,35 +30,74 @@ class AgentCapability(Enum):
     SYNTHESIS = "synthesis"
     EVALUATION = "evaluation"
 
+class TaskCategory(Enum):
+    """Task category for compatibility"""
+    ANALYSIS = "analysis"
+    RESEARCH = "research"
+    CODE = "code"
+    CREATIVE = "creative"
+    SYNTHESIS = "synthesis"
+    EVALUATION = "evaluation"
+
 
 @dataclass
 class Agent:
     """Represents an AI agent with specific capabilities"""
     name: str
-    agent_type: str
-    capabilities: List[AgentCapability]
-    max_complexity: int  # Maximum complexity it can handle (1-10)
+    capabilities: List[Any] = field(default_factory=list)
+    agent_type: str = "general"
+    max_complexity: int = 10
     current_load: int = 0
     success_rate: float = 0.95  # Historical success rate
-
-    def can_handle(self, task_type: TaskType, complexity: int) -> bool:
+    type: str = ""
+    available: bool = True
+
+    def __post_init__(self):
+        if self.type:
+            self.agent_type = self.type
+        elif self.agent_type:
+            self.type = self.agent_type
+
+    def can_handle(self, task_type: Any, complexity: Any) -> bool:
         """Check if agent can handle the task"""
-        # Check complexity
+        if isinstance(complexity, float):
+            complexity = int(complexity * 10)
+        else:
+            complexity = int(complexity)
+
         if complexity > self.max_complexity:
             return False
-
-        # Check capability match
-        capability_map = {
-            TaskType.ANALYSIS: AgentCapability.ANALYSIS,
-            TaskType.RESEARCH: AgentCapability.RESEARCH,
-            TaskType.CODE: AgentCapability.CODE_GENERATION,
-            TaskType.CREATIVE: AgentCapability.CREATIVE,
-            TaskType.SYNTHESIS: AgentCapability.SYNTHESIS,
-            TaskType.EVALUATION: AgentCapability.EVALUATION,
-        }
-
-        required_capability = capability_map.get(task_type, AgentCapability.ANALYSIS)
-        return required_capability in self.capabilities
+
+        # Support both string and Enum for task_type
+        task_str = task_type.value if hasattr(task_type, "value") else str(task_type)
+
+        # Support both string and Enum for capabilities
+        agent_caps = []
+        for cap in self.capabilities:
+            if hasattr(cap, "value"):
+                agent_caps.append(cap.value)
+                if cap == AgentCapability.CODE_GENERATION:
+                    agent_caps.append("code")
+            else:
+                agent_caps.append(str(cap))
+                if str(cap) == "code":
+                    agent_caps.append("code_generation")
+
+        # Map task type to capability
+        required_caps = [task_str]
+        if task_str == "code":
+            required_caps.append("code_generation")
+        elif task_str == "code_generation":
+            required_caps.append("code")
+        elif task_str == "analytical":
+            required_caps.append("analysis")
+        elif task_str == "analysis":
+            required_caps.append("analytical")
+        elif task_str == "creative":
+            required_caps.append("brainstorm")
+            required_caps.append("ideate")
+
+        return any(rc in agent_caps for rc in required_caps)
 
     def get_load_score(self) -> float:
         """Get current load score (0-1, higher = busier)"""
@@ -177,13 +216,24 @@ def __init__(self, registry: Optional[AgentRegistry] = None):
         """
         self.registry = registry or AgentRegistry()
         logger.info("AgentRouter initialized")
+
+    def register_agent(self, agent: Agent):
+        self.registry.add_agent(agent)
+
+    @property
+    def agent_registry(self) -> Dict[str, Agent]:
+        return self.registry.agents
 
     def select_best_agent(
         self,
-        task_type: TaskType,
-        complexity: int,
+        task_type: Any,
+        complexity: Any,
         preferences: Optional[List[str]] = None
     ) -> Optional[Agent]:
+        if isinstance(complexity, float):
+            complexity = int(complexity * 10)
+        else:
+            complexity = int(complexity)
         """
         Select the best agent for a task.
 
@@ -200,8 +250,9 @@ def select_best_agent(
         Returns:
             Selected Agent, or None if no suitable agent found
         """
+        task_str = task_type.value if hasattr(task_type, "value") else str(task_type)
         logger.debug(
-            f"Selecting agent for {task_type.value} task "
+            f"Selecting agent for {task_str} task "
             f"(complexity={complexity})"
         )
 
@@ -218,18 +269,22 @@ def select_best_agent(
 
         # Then, find all capable agents
         for agent in self.registry.list_agents():
+            if not getattr(agent, "available", True):
+                continue
             if agent.can_handle(task_type, complexity):
                 # Calculate score: lower is better
                 # Lower load is better, higher success rate is better
                 score = (
                     agent.get_load_score() * 0.7 +  # 70% weight on load
                     (1 - agent.success_rate) * 0.3   # 30% weight on success rate
                 )
+                if agent.agent_type == task_str or agent.type == task_str:
+                    score -= 1.0 # Significant bonus for exact type matching!
                 candidates.append((agent, score))
 
         if not candidates:
             logger.warning(
-                f"No suitable agent found for {task_type.value} "
+                f"No suitable agent found for {task_str} "
                 f"(complexity={complexity})"
             )
             return None

diff --git a/agents/iterative_agent.py b/agents/iterative_agent.py
@@ -9,8 +9,12 @@
 from typing import Dict, Any, Optional
 import logging
 
-from .chain_of_thought_agent import ChainOfThoughtAgent
-from .output_validator import OutputValidator
+try:
+    from .chain_of_thought_agent import ChainOfThoughtAgent
+    from .output_validator import OutputValidator
+except (ImportError, ValueError):
+    from chain_of_thought_agent import ChainOfThoughtAgent
+    from output_validator import OutputValidator
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)

diff --git a/agents/meta_learner.py b/agents/meta_learner.py
@@ -268,9 +268,9 @@ def _classify_task(self, task: str) -> str:
         task_lower = task.lower()
 
         keywords = {
+            'coding': ['code', 'algorithm', 'program', 'function', 'class'],
             'analytical': ['analyze', 'compare', 'evaluate', 'explain cause'],
             'creative': ['create', 'generate', 'imagine', 'write', 'design'],
-            'coding': ['code', 'algorithm', 'program', 'function', 'class'],
             'explanation': ['what', 'how', 'why', 'explain', 'describe'],
             'evidence': ['prove', 'support', 'evidence', 'data']
         }

diff --git a/agents/model_router.py b/agents/model_router.py
@@ -137,8 +137,8 @@ def select_model(self, category: str) -> str:
         In simple version, returns first available model.
         """
         if category not in MODEL_PROFILES:
-            logger.warning(f"Unknown category {category}, using general")
-            category = "general"
+            logger.warning(f"Unknown category {category}, using default model")
+            return DEFAULT_MODEL
 
         models = MODEL_PROFILES[category]["models"]
         selected = models[0] if models else DEFAULT_MODEL
@@ -194,9 +194,6 @@ def route_and_execute(
 
     def get_routing_stats(self) -> Dict[str, Any]:
         """Get statistics on routing decisions."""
-        if not self.routing_history:
-            return {"total_tasks": 0}
-
         categories_used = {}
         models_used = {}
 
@@ -215,7 +212,7 @@ def get_routing_stats(self) -> Dict[str, Any]:
                 sum(c["confidence"] for c in self.classification_history)
                 / max(len(self.classification_history), 1),
                 2,
-            ),
+            ) if self.classification_history else 0.0,
         }
 
 

diff --git a/agents/performance_optimizer.py b/agents/performance_optimizer.py
@@ -206,7 +206,16 @@ def profile_task(
     def get_performance_summary(self) -> Dict[str, Any]:
         """Get performance summary"""
         if not self.metrics:
-            return {'metrics_count': 0}
+            return {
+                'total_executions': 0,
+                'cache_hits': 0,
+                'cache_hit_rate': 0,
+                'avg_execution_time': 0,
+                'min_execution_time': 0,
+                'max_execution_time': 0,
+                'cache_size': len(self.query_cache),
+                'metrics_count': 0
+            }
 
         execution_times = [m.execution_time for m in self.metrics]
         cache_hits = sum(1 for m in self.metrics if m.cache_hit)
@@ -218,7 +227,8 @@ def get_performance_summary(self) -> Dict[str, Any]:
             'avg_execution_time': sum(execution_times) / len(execution_times) if execution_times else 0,
             'min_execution_time': min(execution_times) if execution_times else 0,
             'max_execution_time': max(execution_times) if execution_times else 0,
-            'cache_size': len(self.query_cache)
+            'cache_size': len(self.query_cache),
+            'metrics_count': len(self.metrics)
         }
 
     def optimize_allocation(self) -> Dict[str, Any]:

diff --git a/agents/result_synthesizer.py b/agents/result_synthesizer.py
@@ -46,8 +46,10 @@ def __init__(self, cot_agent=None, validator=None):
     def synthesize(
         self,
         main_task: str,
-        subtask_results: Dict[str, TaskResult],
-        subtask_descriptions: Dict[str, str]
+        subtask_results: Optional[Dict[str, Any]] = None,
+        subtask_descriptions: Optional[Dict[str, str]] = None,
+        results: Optional[Dict[str, Any]] = None,
+        task_descriptions: Optional[Dict[str, str]] = None
     ) -> str:
         """
         Synthesize multiple sub-task results into final answer.
@@ -67,15 +69,36 @@ def synthesize(
         Returns:
             Final synthesized answer
         """
+        if subtask_results is None:
+            subtask_results = results or {}
+        if subtask_descriptions is None:
+            subtask_descriptions = task_descriptions or {}
+
+        # Normalize subtask_results to always have TaskResult objects
+        normalized_results = {}
+        for k, v in subtask_results.items():
+            if isinstance(v, dict):
+                ans = v.get("answer") or v.get("analysis") or v.get("result") or v.get("feedback") or ""
+                confidence = v.get("confidence", 0.9)
+                quality = v.get("quality") or (confidence * 5.0)
+                normalized_results[k] = TaskResult(
+                    task_id=k,
+                    status=TaskStatus.COMPLETED,
+                    result=ans,
+                    quality_score=quality
+                )
+            else:
+                normalized_results[k] = v
+
         logger.info(
-            f"Synthesizing {len(subtask_results)} subtask results "
+            f"Synthesizing {len(normalized_results)} subtask results "
             f"for main task: {main_task[:80]}..."
         )
 
         # Step 1: Filter successful results
         successful_results = {
             task_id: result
-            for task_id, result in subtask_results.items()
+            for task_id, result in normalized_results.items()
             if result.status == TaskStatus.COMPLETED and result.result
         }
 

diff --git a/agents/task_decomposer.py b/agents/task_decomposer.py
@@ -66,7 +66,7 @@ class TaskDecomposition:
     """Result of task decomposition"""
     main_task: str
     main_task_type: TaskType
-    complexity: int
+    complexity: float
     subtasks: List[SubTask]
     execution_layers: List[List[str]]  # Groups of tasks that can run in parallel
     estimated_total_time: float  # minutes
@@ -209,7 +209,7 @@ def decompose(
         result = TaskDecomposition(
             main_task=task,
             main_task_type=main_task_type,
-            complexity=main_complexity,
+            complexity=main_complexity / 10.0,
             subtasks=subtasks,
             execution_layers=execution_layers,
             estimated_total_time=total_time,
@@ -242,7 +242,9 @@ def _estimate_complexity(self, task: str) -> int:
 
         Uses keyword matching, length, and task type indicators.
         """
-        complexity = 1
+        if len(task.split()) < 5:
+            return 1
+        complexity = 3
         task_lower = task.lower()
 
         # Base complexity from indicators
-Original file line number
+Diff line change
@@ Expand Up @@
     - [**simplified-stack**](simplified-stack/README.ca.md): Versió lleugera optimitzada per al desenvolupament local que integra Drupal, Obsidian i Forgejo per a fluxos de treball d'IA aïllats.
     - [**very-simplified-stack**](very-simplified-stack/README.ca.md): Versió minimalista que elimina l'orquestració de n8n i es centra en serveis de veu i l'API d'agent Cognito, dissenyada per connectar amb una instància d'Ollama externa.
+    - [**AGI Agents & NOOA Framework**](agents/README.md): Mòdul d'agents cognitius que implementa el Roadmap de 5 fases (Chain-of-Thought, Autovalidació i iteració, Memòria a llarg termini amb SQLite + Vectorial, Multi-agent, i Autonomia) integrat de manera nativa amb suport per al paradigma NOOA (NVIDIA-labs Object Oriented Agents).
     ---
@@ Expand Down @@