1. Define approved routes
Keep a short model allowlist per workflow and require review before adding a more expensive or experimental route.
Cost controls
Set LLM cost guardrails before scaling API traffic
Use allowlists, token caps, route reviews and usage logs around https://www.tken.shop/v1 so experiments do not silently become uncontrolled production spend.
Node.js
model allowlist
const allowedModels = new Set(
process.env.TKEN_ALLOWED_MODELS.split(",").map(value => value.trim())
);
function assertAllowedModel(model) {
if (!allowedModels.has(model)) {
throw new Error(`Model is not approved for this workflow: ${model}`);
}
}
const model = process.env.TKEN_CHAT_MODEL;
assertAllowedModel(model);
await client.chat.completions.create({
model,
max_tokens: 300,
messages: [{ role: "user", content: "Answer briefly." }]
});Cap output tokens per workflow
Review model routes before scaling
Track usage without storing secrets
Usage log shape
Store operational metadata that helps control spend, but avoid keeping raw prompts, customer secrets, authorization headers or private documents in routine logs.
Safe metadata
example record
{
"workflow": "support_summary",
"model": "approved-model-id",
"base_url": "https://www.tken.shop/v1",
"status": 200,
"latency_ms": 1840,
"prompt_tokens": 420,
"completion_tokens": 96,
"fallback_used": false
}Cost guardrail checklist
2. Cap request size
Limit max output tokens, trim repeated context and reject inputs that exceed the workflow's expected size.
3. Review usage before raising traffic
Check error rate, fallback rate, token usage and live pricing before moving from tests to larger production batches.
Disclosure: TKEN is an independent third-party API gateway. Use the
live pricing and account views for current model availability and
cost before scaling.
Start with measured traffic
Route small tests through https://www.tken.shop/v1, review usage, then raise limits only after the economics are clear.