Skip to content

INSTALLATION.md

Latest commit

 

History

History
197 lines (155 loc) · 5.79 KB

INSTALLATION.md

File metadata and controls

197 lines (155 loc) · 5.79 KB
Raw

MultiMind SDK Installation Guide

Quick Start

Basic Installation (Core Only)

pip install multimind-sdk

This installs the core: multi-model chat (OpenAI, Claude, Ollama), routing, CLI, and context transfer. Ollama needs no extra — the client talks to a running Ollama instance over HTTP.

Optional Features (Install What You Need)

These are the extras actually defined in pyproject.toml:

RAG Support

pip install "multimind-sdk[rag]"

Adds: faiss-cpu, sentence-transformers, beautifulsoup4, lxml, numpy

Additional Vector Store Backends

pip install "multimind-sdk[vector-stores]"

Adds: chromadb, qdrant-client, weaviate-client, pinecone-client, pymilvus, elasticsearch

Agent Framework

pip install "multimind-sdk[agents]"

Adds: the [memory] extra (redis, numpy) that agents rely on

Memory Backends

pip install "multimind-sdk[memory]"

Adds: redis, numpy

Document Processing

pip install "multimind-sdk[documents]"

Adds: pdfplumber, PyPDF2, python-docx, python-pptx, pillow, pytesseract, unstructured

Fine-tuning (CPU / general)

pip install "multimind-sdk[finetune]"

Adds: torch, transformers, datasets, accelerate, peft, scikit-learn, optuna. Also required for the non-transformer models (Mamba, RWKV), which are HuggingFace-backed.

Fine-tuning (GPU, Linux + CUDA only)

pip install "multimind-sdk[finetune-gpu]"

Adds: everything in [finetune] plus bitsandbytes (Linux only; does not install on macOS/ARM)

Compliance Features

pip install "multimind-sdk[compliance]"

Adds: cryptography, bcrypt, pycryptodome, plotly, dash, pandas (dashboard visualization included)

Gateway / API Server

pip install "multimind-sdk[gateway]"

Adds: fastapi, uvicorn, redis, PyJWT, python-multipart

MCP Server (Claude Desktop / Claude Code integration)

pip install "multimind-sdk[mcp]"

Adds: mcp (the Model Context Protocol SDK) — needed to run the compliance MCP server (python -m multimind.mcp_server) described in docs/mcp-server.md

Framework Integrations

pip install "multimind-sdk[langchain]"    # langchain-core
pip install "multimind-sdk[llamaindex]"   # llama-index-core
pip install "multimind-sdk[crewai]"       # crewai

Each adds only the corresponding framework's core package, for the adapters described in docs/integrations.md.

Everything

pip install "multimind-sdk[all]"

All of the above except dev, finetune-gpu, and the framework extras (langchain, llamaindex, crewai) — large download; includes torch and ML frameworks.

Note: extras like [router], [llm], [minimal], and [fine-tuning] documented in older versions of this guide do not exist. The router and LLM providers are part of the core install.

Development Setup

Clone & Install for Development

git clone https://github.com/multimindlab/multimind-sdk.git
cd multimind-sdk
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install in editable mode with dev tools
pip install -e ".[dev]"

Run Tests

pytest
pytest -v --cov=multimind  # With coverage report

Code Quality

# Lint and format (ruff handles both)
ruff check multimind/
ruff format multimind/

# Type checking
mypy multimind/

See the Makefile for shortcuts (make lint, make format, make test, ...).

Dependency Matrix

Extra Key Dependencies
core (no extra) openai, anthropic, httpx, pydantic, pydantic-settings, python-dotenv, click, rich, PyYAML, aiohttp, requests, tenacity, coloredlogs
rag faiss-cpu, sentence-transformers, beautifulsoup4, lxml, numpy
vector-stores chromadb, qdrant-client, weaviate-client, pinecone-client, pymilvus, elasticsearch
agents multimind-sdk[memory]
memory redis, numpy
documents pdfplumber, PyPDF2, python-docx, python-pptx, pillow, pytesseract, unstructured
finetune torch, transformers, datasets, accelerate, peft, scikit-learn, optuna
finetune-gpu multimind-sdk[finetune] + bitsandbytes (Linux only)
compliance cryptography, bcrypt, pycryptodome, plotly, dash, pandas
gateway fastapi, uvicorn, redis, PyJWT, python-multipart
mcp mcp
langchain langchain-core
llamaindex llama-index-core
crewai crewai
dev pytest, ruff, mypy, pre-commit, bandit, pip-audit, build, twine, sphinx

Troubleshooting

Issue: ModuleNotFoundError for specific feature

Solution: Install the corresponding extra:

pip install "multimind-sdk[extra-name]"

Issue: CUDA/GPU support for PyTorch

Solution: Install the GPU build of torch first, then the extra:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install "multimind-sdk[finetune]"

Issue: bitsandbytes fails to install on macOS/ARM

Solution: bitsandbytes is Linux/CUDA-only. Use [finetune] instead of [finetune-gpu] on macOS.

Verification

Test your installation:

import multimind
print(multimind.__version__)

# Core model wrappers (no extras needed)
from multimind import OpenAIModel, ClaudeModel
from multimind.models.ollama import OllamaModel
print("Core models import OK")

# Optional features (if installed)
try:
    from multimind.vector_store.base import VectorStoreFactory
    print("Vector store module OK")
except ImportError:
    print("Vector stores not installed - run: pip install 'multimind-sdk[rag]'")

Next Steps

  1. Check out the Getting Started Guide
  2. Browse Examples
  3. Read the API Reference
  4. Join our Discord Community