Python SDK

pip install nyraxis-sdk

Initialization

import nyraxis_sdk

nyraxis_sdk.init(
    api_key="nyx_...",              # Required
    base_url="https://api.nyraxis.io",  # Default
    enforce=True,                   # Auto-enforce mode
    instrument=False,               # OpenTelemetry tracing
    agent_name="my-agent",          # Agent identifier
)

Parameter	Default	Description
`api_key`	—	Your API key (starts with `nyx_`)
`base_url`	`https://api.nyraxis.io`	API endpoint (change for self-hosted)
`enforce`	`False`	Enable auto-enforce on OpenAI/Anthropic SDKs
`instrument`	`False`	Enable OpenTelemetry auto-instrumentation
`agent_name`	`None`	Name for this agent in the dashboard
`agent_version`	`None`	Version string for tracking

Evaluate

result = nyraxis_sdk.evaluate(
    input="User message to check",
    output=None,                    # Optional: model output
    policies=["pii_detection"],     # Optional: specific policies
    mode="thorough",                # "thorough" or "fast"
)

result.allowed      # bool
result.violations   # list[Violation]
result.latency_ms   # float

Violation object

@dataclass
class Violation:
    policy_type: str      # e.g. "pii_detection"
    severity: str         # "low" | "medium" | "high" | "critical"
    description: str      # Human-readable explanation
    details: dict         # Provider-specific metadata

Auto-Enforce

When enforce=True, the SDK patches supported LLM SDKs so every call is evaluated before reaching the model.

import nyraxis_sdk
import openai

nyraxis_sdk.init(api_key="nyx_...", enforce=True)

client = openai.OpenAI()
try:
    client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "malicious input"}],
    )
except nyraxis_sdk.NyraxisBlockedError as e:
    print(e.violations)  # list of violations that triggered the block

Enforce options

Parameter	Default	Description
`enforce_fail_open`	`True`	Pass through if Nyraxis is unreachable
`enforce_timeout_s`	`5.0`	HTTP timeout per evaluation
`enforce_cache_ttl_s`	`60.0`	Cache identical inputs (0 to disable)

Supported SDKs

openai (v1+)
anthropic
langchain / langchain-openai
Any HTTP-based LLM client (via manual evaluate)

OpenTelemetry Instrumentation

nyraxis_sdk.init(
    api_key="nyx_...",
    instrument=True,
    agent_name="research-agent",
)
# All LLM calls auto-traced — no code changes needed

Traces are sent to {base_url}/api/v1/ingest/otel.

Low-level Client

from nyraxis_sdk import NyraxisClient

client = NyraxisClient(api_key="nyx_...", base_url="https://api.nyraxis.io")
result = client.evaluate(input="some text")
client.close()

Shutdown

nyraxis_sdk.shutdown()

Closes the HTTP client and releases resources. Call this when your application exits.

Python SDK

On this page