This quickstart uses proxy mode because it is the fastest path to coverage. If you need to gate non-LLM actions too, continue with the Xenovia Python SDK after this.
1. Collect the two values you need
From the Xenovia platform:
- Create a proxy and note the proxy ID.
- Generate an API key (
xe_...) scoped to that proxy.
Set them as environment variables:
export XENOVIA_API_KEY=xe_...
export XENOVIA_PROXY_ID=your-proxy-id
2. Point your client at Xenovia
Python (OpenAI SDK)
Node.js (OpenAI SDK)
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["XENOVIA_API_KEY"],
base_url=f"https://runtime.xenovia.io/a/{os.environ['XENOVIA_PROXY_ID']}/openai/v1"
)
import OpenAI from "openai"
const client = new OpenAI({
apiKey: process.env.XENOVIA_API_KEY,
baseURL: `https://runtime.xenovia.io/a/${process.env.XENOVIA_PROXY_ID}/openai/v1`
})
3. Send your first governed request
response = client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "Summarise AI governance in one sentence."}]
)
print(response.choices[0].message.content)
const response = await client.chat.completions.create({
model: "gpt-4o-mini",
messages: [{ role: "user", content: "Summarise AI governance in one sentence." }]
})
console.log(response.choices[0].message.content)
4. Add a stable session ID
Group related calls into a session so multi-turn conversations appear together in Traces.
import uuid
client = OpenAI(
api_key=os.environ["XENOVIA_API_KEY"],
base_url=f"https://runtime.xenovia.io/a/{os.environ['XENOVIA_PROXY_ID']}/openai/v1",
default_headers={"X-Xenovia-Session-Id": str(uuid.uuid4())}
)
import { randomUUID } from "crypto"
const client = new OpenAI({
apiKey: process.env.XENOVIA_API_KEY,
baseURL: `https://runtime.xenovia.io/a/${process.env.XENOVIA_PROXY_ID}/openai/v1`,
defaultHeaders: { "X-Xenovia-Session-Id": randomUUID() }
})
The session ID must be a valid UUID. Xenovia validates the format and returns a 400 if it is malformed.
5. Confirm the trace exists
After the request completes, check the platform and confirm:
- A trace exists for the request
- The trace shows the expected model and proxy
- The response includes
X-Xenovia-Session-Id and X-Xenovia-Trace-Id
6. Attach a first policy
Start with one policy that proves enforcement is working. This example blocks a destructive tool name:
package xenovia.policy
default allow = true
deny if {
some tool in input.tool_names
tool == "delete_database"
}
7. Handle a policy block
When a request is blocked, the runtime returns 403 Forbidden. The OpenAI SDK raises this as PermissionDeniedError.
from openai import PermissionDeniedError
try:
response = client.chat.completions.create(
model="gpt-4o-mini",
messages=[{"role": "user", "content": "Delete the database."}],
tools=[{
"type": "function",
"function": {
"name": "delete_database",
"description": "Delete all records",
"parameters": {"type": "object", "properties": {}}
}
}]
)
except PermissionDeniedError as e:
print(f"Blocked by policy: {e.message}")
import { PermissionDeniedError } from "openai"
try {
const response = await client.chat.completions.create({
model: "gpt-4o-mini",
messages: [{ role: "user", content: "Delete the database." }],
tools: [{
type: "function",
function: {
name: "delete_database",
description: "Delete all records",
parameters: { type: "object", properties: {} }
}
}]
})
} catch (e) {
if (e instanceof PermissionDeniedError) {
console.log("Blocked by policy:", e.message)
}
}
client = OpenAI(
api_key=os.environ["XENOVIA_API_KEY"],
base_url=f"https://runtime.xenovia.io/a/{os.environ['XENOVIA_PROXY_ID']}/openai/v1",
default_headers={
"X-Xenovia-Session-Id": str(uuid.uuid4()),
"X-Xenovia-Property-environment": "dev",
"X-Xenovia-Property-workflow": "docs-quickstart"
}
)
const client = new OpenAI({
apiKey: process.env.XENOVIA_API_KEY,
baseURL: `https://runtime.xenovia.io/a/${process.env.XENOVIA_PROXY_ID}/openai/v1`,
defaultHeaders: {
"X-Xenovia-Session-Id": randomUUID(),
"X-Xenovia-Property-environment": "dev",
"X-Xenovia-Property-workflow": "docs-quickstart"
}
})
What to do next
After this first pass, move on to:
If your real workflow performs risky actions after the model call, add the Xenovia Python SDK so those actions are governed too. 