requirements.txt

# ─────────────────────────────────────────────────────────────────────────────
# Flowgent — Discovery Agent (Agent 3, Levels 1–3)
# Pinned, minimal dependency set. No PyTorch / TensorFlow / Tesseract / Docker.
# Total install footprint ~60 MB.
# ─────────────────────────────────────────────────────────────────────────────

# --- LLM provider ---
groq>=0.11.0,<1.0.0              # Groq API client (chat + vision)

# --- Level 3: generated connectors + validation (runs their unit tests) ---
requests>=2.31.0,<3.0.0         # HTTP client used by every generated connector

# --- Document parsing ---
PyMuPDF>=1.24.0,<2.0.0           # PDF parsing (C-based, fast, per-page text)
python-docx>=1.1.0,<2.0.0        # .docx parsing (paragraphs + tables)
openpyxl>=3.1.0,<4.0.0           # .xlsx parsing (multi-sheet, computed values)
beautifulsoup4>=4.12.0,<5.0.0    # .html / .htm parsing
Pillow>=10.0.0,<12.0.0           # image normalisation for the vision model
charset-normalizer>=3.3.0,<4.0.0 # robust text encoding detection

# --- Chunking ---
langchain-text-splitters>=0.2.0,<1.0.0  # RecursiveCharacterTextSplitter only

# --- Validation & config ---
pydantic>=2.7.0,<3.0.0           # strict schema validation of LLM output
python-dotenv>=1.0.0,<2.0.0      # load GROQ_API_KEY from .env

# --- CLI / observability ---
rich>=13.7.0,<14.0.0             # tables, progress bars, colour console

# --- Testing (dev) ---
pytest>=8.0.0,<9.0.0