API Reference

AIGuard exposes both a Typer-based CLI and a small Python SDK.

CLI

Command	Purpose
`aiguard project init`	Create `aiguard.yaml` in the current directory.
`aiguard evaluate`	Run all enabled evaluation modules.
`aiguard evaluate adversarial`	Run adversarial attack suite only.
`aiguard evaluate hallucination`	Run hallucination detection only.
`aiguard monitor`	Start the monitoring API + UI.
`aiguard pipeline`	Start the background evaluation pipeline.
`aiguard dev`	Start pipeline + monitoring API + UI together.
`aiguard review serve`	Start the human review server.
`aiguard storage migrate`	Apply pending storage migrations.
`aiguard ci github`	Print a GitHub Actions workflow.
`aiguard ci gitlab`	Print a GitLab CI configuration.

SDK

`aiguard.configure(...)`

One-time configuration. Optional — the SDK auto-reads aiguard.yaml.

python

aiguard.configure(
    sampling_rate=1.0,     # 0.0 – 1.0
    queue_max_size=10_000,
    project="my-project",
)

`aiguard.chat(...)`

Drop-in wrapper around litellm.completion.

python

response = aiguard.chat(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
    metadata={"feature": "chat-ui"},
)
# Returns the original litellm response unmodified.

`TraceEvent`

Dataclass passed to every trace handler. Key fields:

trace_id: str
timestamp: datetime
model: str
input: list[Message]
output: Message
latency_ms: int
tokens: TokenUsage
metadata: dict