Skip to content

Getting Started

A linear three-section tutorial. Time budget:

  • <5 minutes to first working output — complete section 1.
  • <30 minutes end-to-end — complete all three sections.

By the end you will have installed zeroth-core in a clean virtualenv, built a minimal governed graph with one agent and one tool, and run that graph both as an embedded library and as a standalone FastAPI service with a real human-in-the-loop approval gate resolved over HTTP.

Sections

  1. Install — pip install zeroth-core, set ANTHROPIC_API_KEY, run examples/00_hello.py, see a real LLM call complete. This is the <5 minute gate.
  2. First graph — build a minimal graph with one agent, one tool, and one LLM call using the examples/quickstart.py helper. Drive it to completion in-process via the RuntimeOrchestrator — the library-embedded path.
  3. Service mode & approval — boot the same graph as a FastAPI service with a HumanApprovalNode, POST a run, and resolve the pending approval via curl (and the Python equivalent).

Both OpenAI (section 2/3) and Anthropic (section 1) are supported through litellm; the tutorials use OPENAI_API_KEY and ANTHROPIC_API_KEY respectively. Any litellm-compatible provider works.