Make Your First Model Request¶

Make one model request with a personal key and keep its Trace ID. This tutorial uses Chat Completions; the Model gateway is the authoritative compatibility contract.

Prerequisites¶

Access to an existing Mission Control inference origin and active Team membership.
keys.user.create to create your own virtual key, or an administrator-provisioned ordinary service key. Human roles alone do not permit inference.
A generation model allowed by catalog, Team allowance and key policy with model.generate. An administrator should configure pricing for finite spending caps.

1. Get a personal key¶

Open Access → Virtual keys → Create key, choose your Team and For me, review the generation operation/model and optional cap, then save the one-time secret. Use a service-account key for backend ownership independent of a person. No Builder Agent is needed. See virtual keys.

In a new scratch application directory, set the inference origin and real secret in your local environment. These examples use fictional values:

export MISSION_CONTROL_URL=https://gateway.example.test
export MISSION_CONTROL_API_KEY=mck_example_not_a_real_key

Keep this directory/environment for the SDK examples. The URL is an origin without /v1; scripts append the path. A fictional key safely returns 401 until replaced.

2. List the models you may use¶

curl --fail-with-body \
  --header "Authorization: Bearer $MISSION_CONTROL_API_KEY" \
  "$MISSION_CONTROL_URL/v1/models"

The response is an OpenAI-style object with a data array. Each entry has a virtual model id and capabilities; select one supporting chat.completions and text input. Discovery consumes no request limit. An illustrative entry is:

{"id":"support","object":"model","owned_by":"mission-control","capabilities":{"protocols":["chat.completions","responses"],"input_modalities":["text"],"streaming":true}}

Set the exact ID you discovered (the name support is an example, not a bundled model):

export MISSION_CONTROL_MODEL=support

3. Send a request¶

Choose one recipe. The SDK scripts also confirm the model is visible before making the request. Each sends one non-streaming completion and prints its returned Trace ID. Omit traceparent for this first call so the gateway generates a fresh Trace; see Tracing to continue an existing one.

curlPythonTypeScript

Generate JSON safely from the selected model and send it. Headers are printed before the body.

python3 - <<'PYCODE' > request.json
import json, os
print(json.dumps({
    "model": os.environ["MISSION_CONTROL_MODEL"],
    "messages": [{"role": "user", "content": "Say hello in one sentence."}], "max_completion_tokens": 128,
}))
PYCODE
curl --fail-with-body --include \
  --header "Authorization: Bearer $MISSION_CONTROL_API_KEY" \
  --header "Content-Type: application/json" \
  --data-binary @request.json \
  "$MISSION_CONTROL_URL/v1/chat/completions"

Install the version range used by the repository's Python integration:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install 'openai>=2.45,<3'

Save as first_model.py:

import os
from openai import OpenAI

client = OpenAI(
    base_url=f"{os.environ['MISSION_CONTROL_URL'].rstrip('/')}/v1",
    api_key=os.environ["MISSION_CONTROL_API_KEY"],
    max_retries=0,
)
model = os.environ["MISSION_CONTROL_MODEL"]
visible = [entry.id for entry in client.models.list()]
print("Visible models:", visible)
if model not in visible:
    raise SystemExit("Selected model is not visible; check Team/key access.")
raw = client.chat.completions.with_raw_response.create(
    model=model,
    messages=[{"role": "user", "content": "Say hello in one sentence."}],
    max_completion_tokens=128,
)
print("Trace ID:", raw.headers.get("x-mission-control-trace-id"))
print("Correlation ID:", raw.headers.get("x-mission-control-correlation-id"))
print(raw.parse().choices[0].message.content)

python first_model.py

Install the SDK version pinned by the gateway package:

npm init -y
npm install openai@7.10.0

Save as first-model.mts. Node.js 24 runs this TypeScript directly:

import OpenAI from "openai";

const origin = process.env.MISSION_CONTROL_URL;
const key = process.env.MISSION_CONTROL_API_KEY;
const model = process.env.MISSION_CONTROL_MODEL;
if (!origin || !key || !model) throw new Error("Set all three MISSION_CONTROL variables.");
const client = new OpenAI({
  baseURL: `${origin.replace(/\/$/, "")}/v1`, apiKey: key, maxRetries: 0,
});
const visible: string[] = [];
for await (const entry of client.models.list()) visible.push(entry.id);
console.log("Visible models:", visible);
if (!visible.includes(model)) throw new Error("Selected model is not visible; check Team/key access.");
const { data, response } = await client.chat.completions.create({
  model,
  messages: [{ role: "user", content: "Say hello in one sentence." }],
  max_completion_tokens: 128,
}).withResponse();
console.log("Trace ID:", response.headers.get("x-mission-control-trace-id"));
console.log("Correlation ID:", response.headers.get("x-mission-control-correlation-id"));
console.log(data.choices[0].message.content);

node first-model.mts

Expected result¶

HTTP 200, a completion in choices[0].message.content, and X-Mission-Control-Trace-Id and X-Mission-Control-Correlation-Id headers. Generated text and usage vary with the selected model. A gateway-generated Trace ID is 32 hexadecimal characters; keep it for verification.

Verification¶

Open Observe → Requests and locate the canonical request using its safe correlation/Trace ID. An exact own-key read does not authorize the whole shared Trace. Team observability.read opens metadata; captured content and export use separate permissions. Canonical decision evidence exists even if the Collector is unavailable. Each physical attempt records usage independently; model discovery is not an inference call.

Failure recovery¶

Symptom	Recovery
`401` / `authentication_failed`	Check the inference URL and key; issue a replacement for a revoked or expired key under Access → Virtual keys. Suspension or deactivation revokes a person's keys: after an administrator reactivates you, issue a new key.
Empty discovery or `403` / `operation_denied`	Ask a Team administrator to review your Team membership, the Team allowance and the key's operation/resource policy; share the safe correlation ID, never the key. See Get the access you need.
`400` / `unsupported_feature`	Remove unsupported fields using the Model gateway contract.
`400` / `capability_unavailable`	Select a model whose eligible deployment supports the requested fields.
`429`	Honor `Retry-After`; ask an administrator about the spend, request, token or concurrency limit.
Trace or decision not visible	Confirm the selected Team and the distinct own-key/Team metadata and separate content permissions.