quickstart.mdx

title	Quickstart
description	Record an agent and score its reliability in under a minute.

Requires uv ≥ 0.5 (and Node ≥ 20 for the dashboard).

Install

git clone https://github.com/abhay-codes07/VOLO.git
cd VOLO
make setup        # installs Python (uv) + JS toolchains, syncs deps
uv run volo --help

Fastest end-to-end try

One command records an example agent and scores it:

uv run volo init examples.calc_agent:run --input '{"a":2,"b":3,"c":4}'

You'll see it record a baseline, replay it against the adversarial scenario suite (deterministic, $0), and print a ship / no-ship verdict with the four reliability metrics.

The primitives

Record a run, then replay it deterministically:

uv run volo record examples.echo_agent:run --out ./.volo/recordings/echo.json
uv run volo sim ./.volo/recordings/echo.json     # deterministic replay, $0

Score an agent against scenarios and get a verdict:

uv run volo run ./.volo/recordings/echo.json --agent examples.echo_agent:run

Optional: a free LLM judge

Faithfulness can be scored by the default heuristic (free, deterministic), local Ollama, or a free OpenAI-compatible API (Groq by default). Put a key in .env:

VOLO_OPENAI_COMPAT_OPT_IN=true
GROQ_API_KEY=gsk_...

uv run volo run <recording> --agent <module:fn> --judge groq

Next: Core concepts →