Getting Started

This guide walks you through installing AIGuard, initialising a project, and running your first adversarial and hallucination evaluations.

Prerequisites

Python ≥ 3.10
An API key for the LLM provider you want to test (OpenAI, Anthropic, etc.)

1. Install

$pip install aiguard-safety

See Installation for optional extras (monitoring, review, huggingface).

2. Initialise a project

$aiguard project init --project my-project

This writes an aiguard.yaml file to the current directory with safe defaults. Open it and configure your target model — see Configuration.

3. Run your first evaluation

$aiguard evaluate --mode quick

AIGuard runs the enabled modules from aiguard.yaml against the configured target model and writes results to .aiguard/aiguard.db (SQLite).

4. Wrap LLM calls with the SDK

Replace direct LLM calls with aiguard.chat(). It's a drop-in wrapper around LiteLLM and emits a non-blocking trace event for every call.

app.pypython

import aiguard

# Optional — auto-reads aiguard.yaml otherwise
aiguard.configure(sampling_rate=1.0)

response = aiguard.chat(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Explain transformers"}],
)
print(response.choices[0].message.content)

5. Start the dev environment

Run the full local stack (pipeline + monitoring API + React UI) in one command:

$aiguard dev

Monitoring API on http://localhost:8080
Monitoring UI on http://localhost:3000