Skip to content

Installation

JAZ is an open-source Python framework for building LLM (AI) agents that write and run their own code: you describe a task, the agent writes Python, runs it, and returns a typed result. (See Concepts for the full mental model.)

  • Python ≥ 3.12

JAZ is published to PyPI as jaz-lang (the import name is jaz):

Terminal window
pip install jaz-lang

Optional extras (for example, distributed tracing):

Terminal window
pip install "jaz-lang[tracing]"
import jaz # the package installs as jaz-lang but imports as jaz

JAZ talks to model providers through LiteLLM by default, so it reads the standard provider environment variables. A model is named by a LiteLLM route like openai/gpt-5-mini or anthropic/claude-sonnet-4-5, and the provider prefix decides which key is needed.

The Quick start and guide examples use an OpenAI model (openai/gpt-5-mini), so set OPENAI_API_KEY to follow along. To use a different provider, configure the model and set that provider’s key instead (see Budgets & configuration). Provide the key via either:

  • Shell profile — add export OPENAI_API_KEY=sk-... to ~/.zshrc or ~/.bashrc (simplest, available globally).
  • .env file — create a .env at your project root with one KEY=value per line, then load it with direnv (dotenv in an .envrc) or your editor’s env support.
Terminal window
export OPENAI_API_KEY=sk-...
export ANTHROPIC_API_KEY=sk-ant-...

Continue to the Quick start to run your first agent — or skip the install entirely and try a Colab notebook.

Stuck on the install? Ask in the JAZ Discord.