Installation - Exo Docs

Requirements

Python 3.11+ — Exo uses modern Python features (types.UnionType, asyncio.TaskGroup, ExceptionGroup) that require Python 3.11 or later.
An LLM API key — At minimum, you need an API key from OpenAI or Anthropic.
UV — Exo uses UV for package management.

Install from Git

Exo is not yet published on PyPI. Install directly from the Git repository:

bash

# Install UV if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone the repository
git clone https://github.com/Midsphere-AI/exo-ai.git && cd exo-ai

# Sync all workspace packages (installs all packages in editable mode)
uv sync

# Verify the installation
uv run python -c "from exo import Agent, run, tool; print('OK')"

Environment Variables

Set API keys for the LLM providers you plan to use:

bash

# OpenAI (required for "openai:gpt-4o", "openai:gpt-4o-mini", etc.)
export OPENAI_API_KEY="sk-..."

# Anthropic (required for "anthropic:claude-sonnet-4-20250514", etc.)
export ANTHROPIC_API_KEY="sk-ant-..."

You can also pass API keys programmatically when constructing a provider:

python

from exo.models.provider import get_provider

provider = get_provider("openai:gpt-4o", api_key="sk-...")

Or set them in a .env file and load with your preferred method (e.g., python-dotenv).

Running Tests

bash

# Run all tests
uv run pytest

# Run tests for a specific package
uv run pytest packages/exo-core/tests/
uv run pytest packages/exo-models/tests/

# Run with verbose output
uv run pytest -v

Linting and Type Checking

bash

# Lint with ruff
uv run ruff check packages/

# Type-check with pyright
uv run pyright packages/exo-core/
uv run pyright packages/exo-models/

Verify Your Installation

Run this minimal script to confirm everything is working:

python

from exo import Agent, tool

@tool
def hello(name: str) -> str:
    """Say hello."""
    return f"Hello, {name}!"

agent = Agent(
    name="greeter",
    model="openai:gpt-4o-mini",
    instructions="You are a friendly greeter.",
    tools=[hello],
)

print(agent.describe())
# {'name': 'greeter', 'model': 'openai:gpt-4o-mini', 'tools': ['hello'],
#  'handoffs': [], 'max_steps': 10, 'output_type': None}

This script only constructs the agent and does not make any LLM calls, so it works without an API key. If this runs without errors, your installation is correct.

Next Steps

Continue to the Quickstart to build and run your first agent.