parishad 0.1.0__py3-none-any.whl

parishad/data/models.json

@@ -0,0 +1,108 @@
+{
+  "version": "0.1.0",
+  "updated": "2024-12-30",
+  "sources": {
+    "ollama": {
+      "name": "Ollama",
+      "icon": "🦙",
+      "color": "#4a9eff",
+      "url": "https://ollama.ai/library",
+      "models": [
+        {"name": "Llama 3.2 1B", "shortcut": "llama3.2:1b", "size_gb": 0.8, "params": "1B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Compact and fast", "tags": ["lightweight", "fast", "chat"]},
+        {"name": "Llama 3.2 3B", "shortcut": "llama3.2:3b", "size_gb": 2.0, "params": "3B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Balanced performance", "tags": ["balanced", "chat", "code"]},
+        {"name": "Llama 3.1 8B", "shortcut": "llama3.1:8b", "size_gb": 5.0, "params": "8B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Meta's flagship model", "tags": ["powerful", "chat", "reasoning"]},
+        {"name": "Llama 3.1 70B", "shortcut": "llama3.1:70b", "size_gb": 40.0, "params": "70B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Largest Llama model", "tags": ["large", "powerful", "reasoning"]},
+        {"name": "Qwen 2.5 0.5B", "shortcut": "qwen2.5:0.5b", "size_gb": 0.4, "params": "0.5B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Ultra-lightweight", "tags": ["tiny", "fast", "efficient"]},
+        {"name": "Qwen 2.5 1.5B", "shortcut": "qwen2.5:1.5b", "size_gb": 1.0, "params": "1.5B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Compact Qwen", "tags": ["lightweight", "fast", "multilingual"]},
+        {"name": "Qwen 2.5 3B", "shortcut": "qwen2.5:3b", "size_gb": 2.0, "params": "3B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Balanced Qwen", "tags": ["balanced", "multilingual", "code"]},
+        {"name": "Qwen 2.5 7B", "shortcut": "qwen2.5:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Excellent reasoning", "tags": ["powerful", "reasoning", "multilingual"]},
+        {"name": "Qwen 2.5 14B", "shortcut": "qwen2.5:14b", "size_gb": 9.0, "params": "14B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Large Qwen model", "tags": ["large", "reasoning", "multilingual"]},
+        {"name": "Qwen 2.5 32B", "shortcut": "qwen2.5:32b", "size_gb": 20.0, "params": "32B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Very large Qwen", "tags": ["very-large", "reasoning", "multilingual"]},
+        {"name": "Qwen 2.5 72B", "shortcut": "qwen2.5:72b", "size_gb": 45.0, "params": "72B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Largest Qwen model", "tags": ["largest", "powerful", "multilingual"]},
+        {"name": "Qwen 2.5 Coder 1.5B", "shortcut": "qwen2.5-coder:1.5b", "size_gb": 1.0, "params": "1.5B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Small coding model", "tags": ["code", "fast", "programming"]},
+        {"name": "Qwen 2.5 Coder 7B", "shortcut": "qwen2.5-coder:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Powerful coding model", "tags": ["code", "programming", "powerful"]},
+        {"name": "Qwen 2.5 Coder 32B", "shortcut": "qwen2.5-coder:32b", "size_gb": 20.0, "params": "32B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Best coding model", "tags": ["code", "programming", "large"]},
+        {"name": "Mistral 7B", "shortcut": "mistral:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Mistral AI", "description": "Fast and capable", "tags": ["fast", "efficient", "chat"]},
+        {"name": "Mistral Small 22B", "shortcut": "mistral-small:22b", "size_gb": 14.0, "params": "22B", "quantization": "Q4_K_M", "distributor": "Mistral AI", "description": "Enterprise-grade", "tags": ["enterprise", "powerful", "reasoning"]},
+        {"name": "Mistral Large 123B", "shortcut": "mistral-large:123b", "size_gb": 75.0, "params": "123B", "quantization": "Q4_K_M", "distributor": "Mistral AI", "description": "Flagship Mistral", "tags": ["flagship", "very-large", "reasoning"]},
+        {"name": "Mixtral 8x7B", "shortcut": "mixtral:8x7b", "size_gb": 26.0, "params": "8x7B", "quantization": "Q4_K_M", "distributor": "Mistral AI", "description": "MoE architecture", "tags": ["moe", "efficient", "powerful"]},
+        {"name": "Mixtral 8x22B", "shortcut": "mixtral:8x22b", "size_gb": 80.0, "params": "8x22B", "quantization": "Q4_K_M", "distributor": "Mistral AI", "description": "Large MoE model", "tags": ["moe", "very-large", "powerful"]},
+        {"name": "Phi-3 Mini", "shortcut": "phi3:mini", "size_gb": 2.5, "params": "3.8B", "quantization": "Q4_K_M", "distributor": "Microsoft", "description": "Compact powerhouse", "tags": ["compact", "efficient", "reasoning"]},
+        {"name": "Phi-3 Medium", "shortcut": "phi3:medium", "size_gb": 8.0, "params": "14B", "quantization": "Q4_K_M", "distributor": "Microsoft", "description": "Medium Phi model", "tags": ["balanced", "reasoning", "math"]},
+        {"name": "Phi-3.5 Mini", "shortcut": "phi3.5:mini", "size_gb": 2.5, "params": "3.8B", "quantization": "Q4_K_M", "distributor": "Microsoft", "description": "Latest Phi mini", "tags": ["compact", "fast", "reasoning"]},
+        {"name": "Gemma 2 2B", "shortcut": "gemma2:2b", "size_gb": 1.5, "params": "2B", "quantization": "Q4_K_M", "distributor": "Google", "description": "Lightweight Gemma", "tags": ["lightweight", "fast", "efficient"]},
+        {"name": "Gemma 2 9B", "shortcut": "gemma2:9b", "size_gb": 5.5, "params": "9B", "quantization": "Q4_K_M", "distributor": "Google", "description": "Balanced Gemma", "tags": ["balanced", "reasoning", "chat"]},
+        {"name": "Gemma 2 27B", "shortcut": "gemma2:27b", "size_gb": 16.0, "params": "27B", "quantization": "Q4_K_M", "distributor": "Google", "description": "Large Gemma model", "tags": ["large", "powerful", "reasoning"]},
+        {"name": "DeepSeek R1 1.5B", "shortcut": "deepseek-r1:1.5b", "size_gb": 1.0, "params": "1.5B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Compact reasoning", "tags": ["reasoning", "compact", "efficient"]},
+        {"name": "DeepSeek R1 7B", "shortcut": "deepseek-r1:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Strong reasoning", "tags": ["reasoning", "powerful", "math"]},
+        {"name": "DeepSeek R1 14B", "shortcut": "deepseek-r1:14b", "size_gb": 9.0, "params": "14B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Large reasoning model", "tags": ["reasoning", "large", "math"]},
+        {"name": "DeepSeek R1 32B", "shortcut": "deepseek-r1:32b", "size_gb": 20.0, "params": "32B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Very large reasoning", "tags": ["reasoning", "very-large", "math"]},
+        {"name": "DeepSeek R1 70B", "shortcut": "deepseek-r1:70b", "size_gb": 45.0, "params": "70B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Largest DeepSeek R1", "tags": ["reasoning", "largest", "math"]},
+        {"name": "DeepSeek Coder V2", "shortcut": "deepseek-coder-v2:16b", "size_gb": 10.0, "params": "16B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Advanced coder", "tags": ["code", "programming", "powerful"]},
+        {"name": "CodeLlama 7B", "shortcut": "codellama:7b", "size_gb": 4.0, "params": "7B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Code-focused Llama", "tags": ["code", "programming", "balanced"]},
+        {"name": "CodeLlama 13B", "shortcut": "codellama:13b", "size_gb": 7.5, "params": "13B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Larger code model", "tags": ["code", "programming", "powerful"]},
+        {"name": "CodeLlama 34B", "shortcut": "codellama:34b", "size_gb": 20.0, "params": "34B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Best CodeLlama", "tags": ["code", "programming", "large"]},
+        {"name": "StarCoder 2 3B", "shortcut": "starcoder2:3b", "size_gb": 2.0, "params": "3B", "quantization": "Q4_K_M", "distributor": "BigCode", "description": "Compact coder", "tags": ["code", "fast", "programming"]},
+        {"name": "StarCoder 2 7B", "shortcut": "starcoder2:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "BigCode", "description": "Balanced coder", "tags": ["code", "balanced", "programming"]},
+        {"name": "StarCoder 2 15B", "shortcut": "starcoder2:15b", "size_gb": 9.0, "params": "15B", "quantization": "Q4_K_M", "distributor": "BigCode", "description": "Large coder", "tags": ["code", "powerful", "programming"]},
+        {"name": "Granite 3 2B", "shortcut": "granite3-dense:2b", "size_gb": 1.5, "params": "2B", "quantization": "Q4_K_M", "distributor": "IBM", "description": "Compact enterprise", "tags": ["enterprise", "compact", "efficient"]},
+        {"name": "Granite 3 8B", "shortcut": "granite3-dense:8b", "size_gb": 5.0, "params": "8B", "quantization": "Q4_K_M", "distributor": "IBM", "description": "Enterprise model", "tags": ["enterprise", "balanced", "reasoning"]},
+        {"name": "Granite Code 3B", "shortcut": "granite-code:3b", "size_gb": 2.0, "params": "3B", "quantization": "Q4_K_M", "distributor": "IBM", "description": "Enterprise coder", "tags": ["code", "enterprise", "programming"]},
+        {"name": "Granite Code 8B", "shortcut": "granite-code:8b", "size_gb": 5.0, "params": "8B", "quantization": "Q4_K_M", "distributor": "IBM", "description": "Large enterprise coder", "tags": ["code", "enterprise", "powerful"]},
+        {"name": "Orca 2 7B", "shortcut": "orca2:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Microsoft", "description": "Reasoning focused", "tags": ["reasoning", "balanced", "efficient"]},
+        {"name": "Orca 2 13B", "shortcut": "orca2:13b", "size_gb": 7.5, "params": "13B", "quantization": "Q4_K_M", "distributor": "Microsoft", "description": "Large reasoning", "tags": ["reasoning", "large", "powerful"]},
+        {"name": "Neural Chat 7B", "shortcut": "neural-chat:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Intel", "description": "Optimized for Intel", "tags": ["chat", "intel", "optimized"]},
+        {"name": "Vicuna 7B", "shortcut": "vicuna:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "LMSYS", "description": "Fine-tuned chatbot", "tags": ["chat", "fine-tuned", "balanced"]},
+        {"name": "Vicuna 13B", "shortcut": "vicuna:13b", "size_gb": 7.5, "params": "13B", "quantization": "Q4_K_M", "distributor": "LMSYS", "description": "Larger chatbot", "tags": ["chat", "fine-tuned", "powerful"]},
+        {"name": "OpenHermes 2.5 7B", "shortcut": "openhermes:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Teknium", "description": "General assistant", "tags": ["chat", "general", "balanced"]},
+        {"name": "Dolphin Mixtral 8x7B", "shortcut": "dolphin-mixtral:8x7b", "size_gb": 26.0, "params": "8x7B", "quantization": "Q4_K_M", "distributor": "Cognitive Computations", "description": "Uncensored MoE", "tags": ["uncensored", "moe", "powerful"]},
+        {"name": "Wizard Coder 15B", "shortcut": "wizard-coder:15b", "size_gb": 9.0, "params": "15B", "quantization": "Q4_K_M", "distributor": "WizardLM", "description": "Wizard coding model", "tags": ["code", "wizard", "powerful"]},
+        {"name": "MathStral 7B", "shortcut": "mathstral:7b", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Mistral AI", "description": "Math specialized", "tags": ["math", "reasoning", "specialized"]},
+        {"name": "Nomic Embed Text", "shortcut": "nomic-embed-text", "size_gb": 0.3, "params": "137M", "quantization": "FP16", "distributor": "Nomic AI", "description": "Text embeddings", "tags": ["embeddings", "text", "rag"]},
+        {"name": "MxBAI Embed Large", "shortcut": "mxbai-embed-large", "size_gb": 0.7, "params": "334M", "quantization": "FP16", "distributor": "MixedBread", "description": "Large embeddings", "tags": ["embeddings", "large", "rag"]}
+      ]
+    },
+    "huggingface": {
+      "name": "HuggingFace",
+      "icon": "🤗",
+      "color": "#ffcc00",
+      "url": "https://huggingface.co/models",
+      "models": [
+        {"name": "Llama 3.2 1B Instruct", "shortcut": "bartowski/Llama-3.2-1B-Instruct-GGUF", "size_gb": 0.8, "params": "1B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Compact instruction model", "tags": ["lightweight", "instruct", "chat"]},
+        {"name": "Llama 3.2 3B Instruct", "shortcut": "bartowski/Llama-3.2-3B-Instruct-GGUF", "size_gb": 2.0, "params": "3B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Balanced instruction model", "tags": ["balanced", "instruct", "chat"]},
+        {"name": "Llama 3.1 8B Instruct", "shortcut": "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF", "size_gb": 5.0, "params": "8B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Flagship instruction model", "tags": ["powerful", "instruct", "chat"]},
+        {"name": "Qwen 2.5 7B Instruct", "shortcut": "Qwen/Qwen2.5-7B-Instruct-GGUF", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Alibaba", "description": "Multilingual GGUF", "tags": ["multilingual", "instruct", "reasoning"]},
+        {"name": "Mistral 7B Instruct v0.3", "shortcut": "bartowski/Mistral-7B-Instruct-v0.3-GGUF", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "Mistral AI", "description": "Latest Mistral instruct", "tags": ["instruct", "fast", "efficient"]},
+        {"name": "Phi-3 Mini 4K Instruct", "shortcut": "microsoft/Phi-3-mini-4k-instruct-gguf", "size_gb": 2.5, "params": "3.8B", "quantization": "Q4", "distributor": "Microsoft", "description": "Compact reasoning", "tags": ["compact", "reasoning", "math"]},
+        {"name": "Phi-3.5 Mini Instruct", "shortcut": "bartowski/Phi-3.5-mini-instruct-GGUF", "size_gb": 2.5, "params": "3.8B", "quantization": "Q4_K_M", "distributor": "Microsoft", "description": "Latest Phi mini", "tags": ["compact", "fast", "reasoning"]},
+        {"name": "Gemma 2 9B IT", "shortcut": "bartowski/gemma-2-9b-it-GGUF", "size_gb": 5.5, "params": "9B", "quantization": "Q4_K_M", "distributor": "Google", "description": "Instruction tuned Gemma", "tags": ["instruct", "balanced", "chat"]},
+        {"name": "DeepSeek R1 Distill Qwen 7B", "shortcut": "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Distilled reasoning", "tags": ["reasoning", "distilled", "efficient"]},
+        {"name": "DeepSeek Coder V2 Lite", "shortcut": "bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF", "size_gb": 10.0, "params": "16B", "quantization": "Q4_K_M", "distributor": "DeepSeek", "description": "Advanced coding", "tags": ["code", "programming", "powerful"]},
+        {"name": "CodeLlama 7B Instruct", "shortcut": "TheBloke/CodeLlama-7B-Instruct-GGUF", "size_gb": 4.0, "params": "7B", "quantization": "Q4_K_M", "distributor": "Meta", "description": "Code instruction", "tags": ["code", "instruct", "programming"]},
+        {"name": "StarCoder2 7B", "shortcut": "QuantFactory/starcoder2-7b-GGUF", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "BigCode", "description": "Open source coder", "tags": ["code", "open-source", "programming"]},
+        {"name": "Zephyr 7B Beta", "shortcut": "TheBloke/zephyr-7b-beta-GGUF", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "HuggingFace", "description": "Fine-tuned Mistral", "tags": ["chat", "fine-tuned", "helpful"]},
+        {"name": "OpenChat 3.5 7B", "shortcut": "TheBloke/openchat_3.5-GGUF", "size_gb": 4.5, "params": "7B", "quantization": "Q4_K_M", "distributor": "OpenChat", "description": "Top-tier chatbot", "tags": ["chat", "conversational", "helpful"]}
+      ]
+    },
+    "lmstudio": {
+      "name": "LM Studio",
+      "icon": "🎨",
+      "color": "#9966ff",
+      "url": "https://lmstudio.ai/models",
+      "models": [
+        {"name": "Llama 3.2 1B Instruct Q4", "shortcut": "lmstudio-community/Llama-3.2-1B-Instruct-GGUF", "size_gb": 0.7, "params": "1B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Optimized for LM Studio", "tags": ["gguf", "optimized", "lightweight"]},
+        {"name": "Llama 3.2 3B Instruct Q4", "shortcut": "lmstudio-community/Llama-3.2-3B-Instruct-GGUF", "size_gb": 1.8, "params": "3B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Balanced GGUF model", "tags": ["gguf", "balanced", "efficient"]},
+        {"name": "Llama 3.1 8B Instruct Q4", "shortcut": "lmstudio-community/Meta-Llama-3.1-8B-Instruct-GGUF", "size_gb": 4.5, "params": "8B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Powerful GGUF model", "tags": ["gguf", "powerful", "versatile"]},
+        {"name": "Qwen 2.5 7B Instruct Q4", "shortcut": "lmstudio-community/Qwen2.5-7B-Instruct-GGUF", "size_gb": 4.3, "params": "7B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Multilingual GGUF", "tags": ["gguf", "multilingual", "reasoning"]},
+        {"name": "Qwen 2.5 14B Instruct Q4", "shortcut": "lmstudio-community/Qwen2.5-14B-Instruct-GGUF", "size_gb": 8.5, "params": "14B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Large Qwen GGUF", "tags": ["gguf", "large", "multilingual"]},
+        {"name": "Mistral 7B Instruct Q4", "shortcut": "lmstudio-community/Mistral-7B-Instruct-v0.3-GGUF", "size_gb": 4.1, "params": "7B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Fast Mistral GGUF", "tags": ["gguf", "fast", "efficient"]},
+        {"name": "Phi-3 Mini Q4", "shortcut": "lmstudio-community/Phi-3-mini-4k-instruct-GGUF", "size_gb": 2.2, "params": "3.8B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Compact Phi GGUF", "tags": ["gguf", "compact", "reasoning"]},
+        {"name": "Phi-3.5 Mini Q4", "shortcut": "lmstudio-community/Phi-3.5-mini-instruct-GGUF", "size_gb": 2.2, "params": "3.8B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Latest Phi GGUF", "tags": ["gguf", "compact", "fast"]},
+        {"name": "Gemma 2 9B IT Q4", "shortcut": "lmstudio-community/gemma-2-9b-it-GGUF", "size_gb": 5.2, "params": "9B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Gemma 2 GGUF", "tags": ["gguf", "balanced", "reasoning"]},
+        {"name": "DeepSeek R1 Distill 7B Q4", "shortcut": "lmstudio-community/DeepSeek-R1-Distill-Qwen-7B-GGUF", "size_gb": 4.3, "params": "7B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Reasoning GGUF", "tags": ["gguf", "reasoning", "distilled"]},
+        {"name": "CodeLlama 7B Instruct Q4", "shortcut": "lmstudio-community/CodeLlama-7b-Instruct-GGUF", "size_gb": 3.8, "params": "7B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Coding GGUF", "tags": ["gguf", "code", "programming"]},
+        {"name": "Qwen 2.5 Coder 7B Q4", "shortcut": "lmstudio-community/Qwen2.5-Coder-7B-Instruct-GGUF", "size_gb": 4.3, "params": "7B", "quantization": "Q4_K_M", "distributor": "LM Studio", "description": "Coding specialist GGUF", "tags": ["gguf", "code", "programming"]}
+      ]
+    }
+  }
+}

parishad/memory/__init__.py

@@ -0,0 +1,79 @@
+"""
+Parishad Memory System.
+Provides vector storage and retrieval capabilities.
+"""
+
+from typing import Any, List, Dict, Optional
+
+class VectorStore:
+    """
+    VectorStore implementation using ChromaDB.
+    """
+    
+    def __init__(self, collection_name: str = "parishad_memory", persist_dir: str = "./.parishad_data/vector_store"):
+        import chromadb
+        import os
+        
+        # Ensure directory exists
+        os.makedirs(persist_dir, exist_ok=True)
+        
+        self.client = chromadb.PersistentClient(path=persist_dir)
+        self.collection = self.client.get_or_create_collection(name=collection_name)
+        
+    def add(self, documents: List[str], metadata: Optional[List[Dict[str, Any]]] = None, ids: Optional[List[str]] = None) -> None:
+        """Add documents to the store."""
+        if not documents:
+            return
+            
+        # Generate IDs if not provided
+        if ids is None:
+            import uuid
+            ids = [str(uuid.uuid4()) for _ in documents]
+            
+        # Ensure metadata is provided for all docs
+        if metadata is None:
+            metadata = [{} for _ in documents]
+            
+        self.collection.add(
+            documents=documents,
+            metadatas=metadata,
+            ids=ids
+        )
+        
+    def query(self, query_text: str, n_results: int = 5) -> List[Dict[str, Any]]:
+        """
+        Query the store.
+        Returns list of dicts with 'content', 'metadata', 'distance'.
+        """
+        try:
+            results = self.collection.query(
+                query_texts=[query_text],
+                n_results=n_results
+            )
+            
+            # Chroma returns lists of lists (one per query)
+            if not results['documents'] or not results['documents'][0]:
+                return []
+                
+            formatted_results = []
+            
+            # Zip the first (and only) query result lists
+            doc_list = results['documents'][0]
+            meta_list = results['metadatas'][0]
+            
+            # Handle distances if available (cosine distance usually)
+            dist_list = results['distances'][0] if results['distances'] else [0.0] * len(doc_list)
+            
+            for doc, meta, dist in zip(doc_list, meta_list, dist_list):
+                formatted_results.append({
+                    'content': doc,
+                    'metadata': meta,
+                    'distance': dist
+                })
+                
+            return formatted_results
+            
+        except Exception as e:
+            # Fallback or empty on error
+            print(f"Vector search error: {e}")
+            return []

parishad/models/__init__.py

@@ -0,0 +1,181 @@
+"""Model abstraction layer for Parishad."""
+
+from .runner import (
+    ModelRunner,
+    ModelConfig,
+    SlotConfig,
+    Backend,
+    ModelRunnerError,
+    UnknownSlotError,
+    ModelBackendError,
+    BackendNotAvailableError,
+    TransformersBackend,
+    OpenAIBackend,
+)
+
+from .backends import (
+    BackendConfig,
+    BackendResult,
+    BackendError,
+    ModelBackend,
+    is_backend_available,
+    get_available_backends,
+)
+
+from .tokenization import (
+    estimate_tokens,
+    estimate_tokens_simple,
+    estimate_tokens_hybrid,
+    estimate_prompt_tokens,
+    count_tokens_tiktoken,
+    is_tiktoken_available,
+    get_tokenizer,
+    register_tokenizer,
+)
+
+from .costs import (
+    estimate_cost,
+    estimate_query_cost,
+    get_model_pricing,
+    ModelPricing,
+    CostMetrics,
+    estimate_flops,
+    get_model_size,
+)
+
+from .profiles import (
+    ProfileManager,
+    ProfileMode,
+    ProfileDefinition,
+    HardwareCapability,
+    EnvironmentInfo,
+    detect_environment,
+    get_default_profile,
+    get_profile_manager,
+    quick_runner,
+    BUILTIN_PROFILES,
+)
+
+# Task 7: Performance optimizations
+from .optimizations import (
+    ResponseCache,
+    PersistentCache,
+    RequestBatcher,
+    ConnectionPool,
+    RateLimiter,
+    OptimizedRunner,
+)
+
+# Task 8: Reliability
+from .reliability import (
+    RetryStrategy,
+    RetryPolicy,
+    TimeoutConfig,
+    TimeoutManager,
+    CircuitState,
+    CircuitBreakerConfig,
+    CircuitBreaker,
+    CircuitOpenError,
+    FallbackChain,
+    HealthStatus,
+    HealthChecker,
+    ResilientBackend,
+)
+
+# Model Download Manager
+from .downloader import (
+    ModelManager,
+    ModelRegistry,
+    ModelInfo,
+    ModelSource,
+    ModelFormat,
+    HuggingFaceDownloader,
+    OllamaManager,
+    LMStudioManager,
+    DownloadProgress,
+    DEFAULT_MODEL_DIR,
+)
+
+
+__all__ = [
+    # Main classes
+    "ModelRunner",
+    "ModelConfig",
+    "SlotConfig",
+    "Backend",
+    # Backend protocol
+    "ModelBackend",
+    "BackendConfig",
+    "BackendResult",
+    # Exceptions
+    "ModelRunnerError",
+    "UnknownSlotError",
+    "ModelBackendError",
+    "BackendNotAvailableError",
+    "BackendError",
+    # Backend implementations
+    "TransformersBackend",
+    "OpenAIBackend",
+    # Backend utilities
+    "is_backend_available",
+    "get_available_backends",
+    # Tokenization
+    "estimate_tokens",
+    "estimate_tokens_simple",
+    "estimate_tokens_hybrid",
+    "estimate_prompt_tokens",
+    "count_tokens_tiktoken",
+    "is_tiktoken_available",
+    "get_tokenizer",
+    "register_tokenizer",
+    # Cost estimation
+    "estimate_cost",
+    "estimate_query_cost",
+    "get_model_pricing",
+    "ModelPricing",
+    "CostMetrics",
+    "estimate_flops",
+    "get_model_size",
+    # Profile management
+    "ProfileManager",
+    "ProfileMode",
+    "ProfileDefinition",
+    "HardwareCapability",
+    "EnvironmentInfo",
+    "detect_environment",
+    "get_default_profile",
+    "get_profile_manager",
+    "quick_runner",
+    "BUILTIN_PROFILES",
+    # Performance optimizations (Task 7)
+    "ResponseCache",
+    "PersistentCache",
+    "RequestBatcher",
+    "ConnectionPool",
+    "RateLimiter",
+    "OptimizedRunner",
+    # Reliability (Task 8)
+    "RetryStrategy",
+    "RetryPolicy",
+    "TimeoutConfig",
+    "TimeoutManager",
+    "CircuitState",
+    "CircuitBreakerConfig",
+    "CircuitBreaker",
+    "CircuitOpenError",
+    "FallbackChain",
+    "HealthStatus",
+    "HealthChecker",
+    "ResilientBackend",
+    # Model Download Manager
+    "ModelManager",
+    "ModelRegistry",
+    "ModelInfo",
+    "ModelSource",
+    "ModelFormat",
+    "HuggingFaceDownloader",
+    "OllamaManager",
+    "LMStudioManager",
+    "DownloadProgress",
+    "DEFAULT_MODEL_DIR",
+]

parishad/models/backends/__init__.py

@@ -0,0 +1,247 @@
+"""
+Backend implementations for Parishad model runners.
+
+This package provides pluggable backend implementations for different LLM providers:
+- LlamaCppBackend: Local GGUF models via llama-cpp-python
+- OpenAIBackend: OpenAI API (and compatible endpoints)
+- OllamaBackend: Ollama local server (OpenAI-compatible)
+- OllamaNativeBackend: Ollama using native API
+- TransformersBackend: HuggingFace Transformers models
+- HuggingFaceBackend: HuggingFace Inference API (cloud)
+- HuggingFaceChatBackend: HuggingFace chat completion API
+- MlxBackend: Apple Silicon (M1/M2/M3/M4) via MLX
+"""
+
+from __future__ import annotations
+
+# Base classes and types - always available
+from .base import (
+    BackendError,
+    BackendConfig,
+    BackendResult,
+    ModelBackend,
+    BaseBackend,
+)
+
+# Conditional imports for optional backends
+_LLAMA_CPP_AVAILABLE = False
+_OPENAI_AVAILABLE = False
+_OLLAMA_AVAILABLE = False
+_OLLAMA_NATIVE_AVAILABLE = False
+_TRANSFORMERS_AVAILABLE = False
+_HUGGINGFACE_AVAILABLE = False
+_MLX_AVAILABLE = False
+
+
+# LlamaCpp
+try:
+    from .llama_cpp import LlamaCppBackend
+    _LLAMA_CPP_AVAILABLE = True
+except ImportError:
+    LlamaCppBackend = None  # type: ignore
+
+# OpenAI
+try:
+    from .openai_api import OpenAIBackend, OllamaBackend
+    _OPENAI_AVAILABLE = True
+    _OLLAMA_AVAILABLE = True
+except ImportError:
+    OpenAIBackend = None  # type: ignore
+    OllamaBackend = None  # type: ignore
+
+# Transformers
+try:
+    from .transformers_hf import TransformersBackend
+    _TRANSFORMERS_AVAILABLE = True
+except ImportError:
+    TransformersBackend = None  # type: ignore
+
+# HuggingFace Inference API
+try:
+    from .huggingface import HuggingFaceBackend, HuggingFaceChatBackend
+    _HUGGINGFACE_AVAILABLE = True
+except ImportError:
+    HuggingFaceBackend = None  # type: ignore
+    HuggingFaceChatBackend = None  # type: ignore
+
+# Ollama Native
+try:
+    from .ollama import OllamaNativeBackend
+    _OLLAMA_NATIVE_AVAILABLE = True
+except ImportError:
+    OllamaNativeBackend = None  # type: ignore
+
+# MLX
+try:
+    from .mlx_lm import MlxBackend
+    _MLX_AVAILABLE = True
+except ImportError:
+    MlxBackend = None  # type: ignore
+
+
+
+
+def is_backend_available(name: str) -> bool:
+    """Check if a backend's dependencies are installed."""
+    if name == "llama_cpp":
+        try:
+            import llama_cpp
+            return True
+        except ImportError:
+            return False
+    elif name == "openai":
+        try:
+            import openai
+            return True
+        except ImportError:
+            return False
+    elif name == "transformers":
+        try:
+            import transformers
+            import torch
+            return True
+        except ImportError:
+            return False
+    elif name == "ollama":
+        try:
+            import openai
+            return True
+        except ImportError:
+            return False
+    elif name == "mlx":
+        try:
+            import mlx_lm
+            return True
+        except ImportError:
+            return False
+
+    elif name == "ollama_native":
+        try:
+            import requests
+            return True
+        except ImportError:
+            return False
+    elif name in ("huggingface", "huggingface_chat"):
+        try:
+            import huggingface_hub
+            return True
+        except ImportError:
+            return False
+    return False
+
+
+def get_available_backends() -> list[str]:
+    """Get a list of currently available backends."""
+    backends = []
+    
+    if is_backend_available("llama_cpp"):
+        backends.append("llama_cpp")
+    if is_backend_available("openai"):
+        backends.append("openai")
+    if is_backend_available("transformers"):
+        backends.append("transformers")
+    if is_backend_available("ollama"):
+        backends.append("ollama")
+    if is_backend_available("mlx"):
+        backends.append("mlx")
+    if is_backend_available("ollama_native"):
+        backends.append("ollama_native")
+    if is_backend_available("huggingface"):
+        backends.append("huggingface")
+        backends.append("huggingface_chat")
+        
+    return backends
+
+
+def get_backend(backend_name: str) -> BaseBackend:
+    """
+    Factory function to get a backend instance by name.
+    
+    Args:
+        backend_name: Name of the backend
+    
+    Returns:
+        Backend instance (not loaded)
+        
+    Raises:
+        ValueError: If backend unknown or deps not installed
+    """
+    backend_map = {
+        "llama_cpp": (LlamaCppBackend, _LLAMA_CPP_AVAILABLE),
+        "openai": (OpenAIBackend, _OPENAI_AVAILABLE),
+        "ollama": (OllamaBackend, _OLLAMA_AVAILABLE),
+        "ollama_native": (OllamaNativeBackend, _OLLAMA_NATIVE_AVAILABLE),
+        "transformers": (TransformersBackend, _TRANSFORMERS_AVAILABLE),
+        "huggingface": (HuggingFaceBackend, _HUGGINGFACE_AVAILABLE),
+        "huggingface_chat": (HuggingFaceChatBackend, _HUGGINGFACE_AVAILABLE),
+        "mlx": (MlxBackend, _MLX_AVAILABLE),
+    }
+    
+    if backend_name not in backend_map:
+        available = ", ".join(backend_map.keys())
+        raise ValueError(
+            f"Unknown backend: '{backend_name}'. "
+            f"Available backends: {available}"
+        )
+    
+    backend_class, is_available = backend_map[backend_name]
+    
+    if not is_available or backend_class is None:
+        raise ValueError(
+            f"Backend '{backend_name}' is not available. "
+            f"Required dependencies are not installed. "
+            f"Available backends: {', '.join(get_available_backends())}"
+        )
+    
+    return backend_class()
+
+
+# Stub backend for testing
+class StubBackend(BaseBackend):
+    """Simple stub backend for testing."""
+    
+    _name = "stub"
+    
+    def load(self, config: BackendConfig) -> None:
+        self._config = config
+        self._model_id = config.model_id
+        self._loaded = True
+    
+    def generate(
+        self,
+        prompt: str,
+        max_tokens: int,
+        temperature: float,
+        top_p: float,
+        stop: list[str] | None = None,
+    ) -> BackendResult:
+        return BackendResult(
+            text="[STUB RESPONSE]",
+            tokens_in=self._estimate_tokens(prompt),
+            tokens_out=10,
+            model_id=self._model_id,
+        )
+
+
+__all__ = [
+    # Base classes
+    "BackendError",
+    "BackendConfig",
+    "BackendResult",
+    "ModelBackend",
+    "BaseBackend",
+    # Backend implementations
+    "LlamaCppBackend",
+    "OpenAIBackend",
+    "OllamaBackend",
+    "OllamaNativeBackend",
+    "TransformersBackend",
+    "HuggingFaceBackend",
+    "HuggingFaceChatBackend",
+    "MlxBackend",
+    "StubBackend",
+    # Utilities
+    "is_backend_available",
+    "get_available_backends",
+    "get_backend",
+]