Nuclia Python Client

In order to install

pip install nuclia

Authentication

Nuclia

You can login with your Nuclia user How to sign-up via

nuclia auth login

Nuclia Knowledgebox

You can login to a specific knowledgebox if you have a Service Token (How to get a Service Token) or your NucliaDB is deployed on-premise

nuclia auth kb --url KB_URL --token SERVICE_TOKEN

KB_URL its the url of the Knowledge Box. On the cloud service you can retrieve it on the dashboard. On the on-premise/community deployment its the url mapped to it.

Nuclia Understanding API

You can login with a Nuclia Understanding API key to process files, predict and train using our system

nuclia auth nua --key ZZZZ

Guardrails

Account guardrail policies can be managed with the synchronous or asynchronous SDK:

from nuclia_models.accounts.guardrails import CreateGuardrailPolicy
from nuclia.sdk import NucliaGuardrails

guardrails = NucliaGuardrails()
policy = guardrails.create(
    CreateGuardrailPolicy(
        name="Safety policy",
        instruction="Flag requests that ask for dangerous instructions.",
        query="Does the request violate the safety policy?",
        enabled=True,
        blocking=True,
    ),
    zone="europe-1",
)

Enabled account policies are evaluated automatically for Predict chat requests. Content can also be evaluated directly against a configured or inline policy:

from nuclia_models.predict.guardrails import GuardrailRequest
from nuclia.sdk import NucliaPredict

result = NucliaPredict().guardrail(
    GuardrailRequest(
        content="Content to evaluate",
        policy_id=str(policy.id),
    )
)

Policy violations always block chat generation. The blocking setting controls failure handling: evaluation failures fail closed when enabled and fail open when disabled.

Documentation

You can find the documentation here

For internal use

Find more information on the internal CI setup here