Skip to content

JAZ Documentation

JAZ is a Python framework for building and optimizing LLM agents through a code-first REPL loop. The LLM generates code, runs it, observes the results, and iterates until the task is complete — so your agent workflows are predictable, auditable, and cost-aware.

from jaz import invoke
from jaz.hooks import ReturnType
result = invoke(ReturnType(int), task="Calculate the factorial of 10")
print(result) # 3628800
  • Code beats prose for workflows. Reasoning expressed as code is more predictable and auditable than natural-language-only agents.
  • Recursive composition. Agents invoke nested agents, each with its own budget.
  • Built-in observability. Logging and OpenTelemetry tracing via composable hooks.
  • Cost control. Enforce cost and call-count budgets across the whole recursion tree.

Prefer to learn by doing? Jump straight into a Colab tutorial.

Questions, ideas, or something that didn’t work? Join the JAZ Discord to ask the people building it, or open an issue on GitHub.