# MIRA - Validated Third-Party Requirements

# Generated by analyzing actual usage across the codebase

# Each package verified to be actively used in non-deprecated files

# == PYTORCH CPU-ONLY INSTALLATION ==

--extra-index-url https://download.pytorch.org/whl/cpu

# == WEB FRAMEWORK & SERVER ==

fastapi # Core web framework - heavily used across API endpoints

starlette # FastAPI dependency - used in middleware

hypercorn # ASGI server with HTTP/2 support - required for all environments

# == DATABASE & DATA PERSISTENCE ==

psycopg[binary] # PostgreSQL driver v3 - postgres_client.py, database utils

psycopg-pool # Connection pooling for psycopg v3

pgvector # PostgreSQL vector extension - postgres_client.py

# == CACHING & KEY-VALUE STORE ==

valkey # Redis-compatible cache - valkey_client.py, many services

# == AUTHENTICATION & SECURITY ==

cryptography # Encryption utilities - credential_security.py, userdata_manager.py

# == AI & MACHINE LEARNING ==

anthropic # Anthropic Claude API SDK - primary LLM provider with prompt caching

openai # OpenAI API client - tools, embeddings (fallback for some tools)

numpy # Numerical computing - embeddings, vector operations, ML utilities

torch # PyTorch (CPU-only via index above) - sentence_transformers.py, bge_embeddings.py

transformers # Hugging Face transformers - embeddings, BGE models

sentence-transformers # Sentence embeddings - mdbr-leaf-ir-asym

optimum[onnxruntime] # ONNX model optimization - sentence_transformers.py, BGE embeddings

onnx # ONNX runtime - sentence_transformers.py, bge_embeddings.py

onnxruntime # ONNX inference engine - bge_embeddings.py

onnxconverter_common # ONNX conversion utilities - bge_embeddings.py

tqdm # Progress bars - bge_embeddings.py model loading

scikit-learn>=1.4.0 # TF-IDF candidate discovery for memory linking pipeline

spacy>=3.7.0 # NER and NLP - entity extraction for memory linking

# Note: Install en_core_web_lg with: python -m spacy download en_core_web_lg

# == SCHEDULING & BACKGROUND TASKS ==

apscheduler # Task scheduling - memory_extraction_service.py, auth_service.py, scheduler_service.py

# == HTTP CLIENT & NETWORKING ==

httpx[http2] # Modern HTTP client with HTTP/2 support - tools, LLM provider, embeddings, weather/web tools

aiohttp # Async HTTP - kasa_tool.py for smart home control

mcp # Model Context Protocol SDK - mcp_client.py for MCP server connections

# == TOOL-SPECIFIC LIBRARIES ==

googlemaps # Google Maps API - maps_tool.py

google-genai # Google Gemini GenAI SDK - imagegen_tool.py

kasa # TP-Link Kasa smart devices - kasa_tool.py

kagiapi # Kagi search API - web_tool.py (primary search provider)

ddgs # DuckDuckGo search - web_tool.py (fallback search provider)

trafilatura # Deterministic web content extraction - web_tool.py (replaces LLM extraction)

hvac # HashiCorp Vault client - vault_client.py

# == BILLING & PAYMENTS ==

# == IMAGE PROCESSING ==

Pillow>=10.0.0 # Image compression - two-tier compression for multimodal messages

# == UTILITIES & HELPERS ==

pydantic # Data validation - widely used across API models and tools

python-dateutil # Date parsing - reminder_tool.py, timezone_utils.py

pytz # Timezone handling - timezone_utils.py

colorama # Terminal colors - logging_config.py

rapidfuzz # Fuzzy string matching - memory deduplication, entity normalization

json-repair # JSON repair for malformed LLM responses - extraction.py

psutil # Process management - playwright_service.py for forced browser shutdown

# == TESTING DEPENDENCIES ==

pytest # Test framework - extensive test suite

pytest-asyncio # Async test support - async tests

# == OPTIONAL DEPENDENCIES ==

# These packages enhance functionality but are not required for core operation

# Install with: pip install python-docx openpyxl playwright

# Document processing - enables richer DOCX/XLSX text extraction

# Falls back to stdlib XML parsing if not installed

python-docx # (optional) DOCX text extraction - document_processing.py

openpyxl # (optional) XLSX text extraction - document_processing.py

# Web scraping - enables JavaScript-rendered page capture

playwright # (optional) Headless browser - web_tool.py

# == PACKAGES TO REVIEW/CONSIDER REMOVING ==

# These packages have limited usage and may be removable:

# - pytz (could potentially use zoneinfo from Python 3.9+ instead)