Skip to content

Installation

Requirements

  • Python 3.11 or later
  • pip or uv package manager

Basic installation

pip install ct-toolkit

This installs CT Toolkit with all core dependencies. No additional packages are needed to get started with OpenAI, Anthropic, or Ollama.

Framework extras

Install optional extras based on the frameworks you use:

pip install "ct-toolkit[langchain]"

Enables TheseusChatModel and TheseusLangChainCallback. Requires langchain-core >= 1.2.

pip install "ct-toolkit[crewai]"

Enables TheseusCrewMiddleware. Requires crewai >= 1.10.

pip install "ct-toolkit[autogen]"

Enables TheseusAutoGenMiddleware. Requires pyautogen >= 0.4.

pip install "ct-toolkit[ml]"

Enables DivergencePenaltyLoss for PyTorch training loops. Requires torch >= 2.0.

pip install "ct-toolkit[langchain,crewai,autogen,ml]"

Development installation

For contributors or if you want to run the test suite:

git clone https://github.com/hakandamar/ct-toolkit
cd ct-toolkit

# Using uv (recommended)
uv sync --all-extras

# Or with pip
pip install -e ".[dev,langchain,crewai,autogen]"

Run the tests:

pytest tests/ -v
# 242 passed, 8 skipped

Always use a virtual environment in production:

python -m venv .venv

# Activate — macOS / Linux
source .venv/bin/activate

# Activate — Windows
.venv\Scripts\activate

pip install ct-toolkit

uv venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
uv add ct-toolkit

conda create -n myenv python=3.11
conda activate myenv
pip install ct-toolkit

Verifying the installation

import ct_toolkit
print(ct_toolkit.__version__)  # e.g. 0.3.3

Or run a quick sanity check:

from ct_toolkit import TheseusWrapper, ConstitutionalKernel

kernel = ConstitutionalKernel.default()
print(f"Kernel: {kernel.name}")
print(f"Anchors: {len(kernel.anchors)}")
print(f"Commitments: {len(kernel.commitments)}")

Expected output:

Kernel: default
Anchors: 4
Commitments: 4

Dependencies overview

CT Toolkit's core dependencies:

Package Purpose
litellm Unified transport layer (OpenAI, Anthropic, Ollama, etc.)
instructor Structured L2/L3 judge responses
numpy L1 embedding calculations
pyyaml Kernel and template YAML loading
pydantic Config validation
cryptography HMAC key management
jinja2 System prompt templating

No heavy ML dependency

CT Toolkit's core does not require PyTorch or any ML framework. The [ml] extra is only needed for training-time DivergencePenaltyLoss.

Next: Quickstart

Build your first protected agent