How to Use EvoLink Smart Router: API Setup and Production Testing

EvoLink Smart Router quick reference

1. Create an API key

Create an EvoLink API key in the dashboard, then store it in an environment variable rather than hardcoding it in application code:

export EVOLINK_API_KEY="your-api-key"

For PowerShell:

$env:EVOLINK_API_KEY="your-api-key"

Keep separate keys for local development, staging, and production so that usage can be reviewed and credentials can be rotated independently.

2. Send your first Smart Router request

curl --request POST \
  --url https://direct.evolink.ai/v1/chat/completions \
  --header "Authorization: Bearer $EVOLINK_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "evolink/auto",
    "messages": [
      {
        "role": "user",
        "content": "Classify this support request as billing, technical, or account access: I cannot sign in after resetting my password."
      }
    ],
    "temperature": 0.2,
    "stream": false
  }'

{
  "id": "chatcmpl-example",
  "object": "chat.completion",
  "model": "actual-routed-model",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "account access"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 26,
    "completion_tokens": 3,
    "total_tokens": 29
  }
}

3. Use Smart Router with Python

If your application already uses the OpenAI Python SDK, point the client at EvoLink's text API base URL and change the model ID:

import os
import time
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["EVOLINK_API_KEY"],
    base_url="https://direct.evolink.ai/v1",
)

started_at = time.perf_counter()

response = client.chat.completions.create(
    model="evolink/auto",
    messages=[
        {
            "role": "user",
            "content": "Summarize this incident report and list the next two engineering actions.",
        }
    ],
    temperature=0.2,
)

latency_ms = round((time.perf_counter() - started_at) * 1000)

print("routed_model:", response.model)
print("latency_ms:", latency_ms)
print("prompt_tokens:", response.usage.prompt_tokens if response.usage else None)
print("completion_tokens:", response.usage.completion_tokens if response.usage else None)
print("output:", response.choices[0].message.content)

Do not log raw prompts or responses if they may contain secrets, personal data, or customer content. Log operational metadata appropriate for your own privacy and retention requirements.

4. Use Smart Router with Node.js

The same integration pattern works with the OpenAI Node.js SDK:

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.EVOLINK_API_KEY,
  baseURL: "https://direct.evolink.ai/v1",
});

const startedAt = performance.now();

const response = await client.chat.completions.create({
  model: "evolink/auto",
  messages: [
    {
      role: "user",
      content: "Extract the customer name, company, priority, and requested deadline as JSON.",
    },
  ],
  temperature: 0,
});

console.log({
  routedModel: response.model,
  latencyMs: Math.round(performance.now() - startedAt),
  promptTokens: response.usage?.prompt_tokens,
  completionTokens: response.usage?.completion_tokens,
  output: response.choices[0]?.message?.content,
});

For structured extraction, validate the returned data in application code. Routing does not remove the need for schema validation, retries, or workflow-specific quality checks.

How EvoLink Smart Router chooses a route

For supported text requests, EvoLink describes the routing flow in five stages:

1. Your application sends an OpenAI-compatible request using evolink/auto.
2. The router evaluates the task type and complexity.
3. The request is mapped to a routing profile such as Fast, Standard, or Reasoning.
4. A suitable candidate model receives the request.
5. The selected model is returned in response.model.

The profiles are task categories, not a permanent public list of models:

Smart Router vs a fixed model

Smart Router is useful when one workflow contains different kinds of text tasks. A fixed model is better when the model itself is part of the product contract.

A practical production architecture often uses both:

• evolink/auto for mixed or evolving text workloads
• explicit model IDs for evaluated, model-specific, or tightly controlled paths

What to log for every routed request

Routing becomes useful only when you can compare its results with a baseline. At minimum, capture:

The quality result can be a deterministic validator, a human review label, a test-suite result, or another task-appropriate evaluation. Avoid reducing every workflow to one generic quality score.

How to test Smart Router before production

Do not switch all traffic after one successful request. Use a staged evaluation:

Step 1: Build a representative fixture set

Collect real examples from the workflow you plan to route. Include normal requests, ambiguous inputs, long prompts, malformed inputs, and cases where a wrong answer creates meaningful product risk.

Step 2: Choose a fixed-model baseline

Run the same fixture set against the fixed model your application currently uses. Keep prompts, parameters, and evaluation rules consistent.

Step 3: Run the Smart Router path

Step 4: Compare by workflow, not by average alone

A router can perform well overall and still be unsuitable for one high-risk feature. Review results by task type, customer tier, latency requirement, and failure impact.

Step 5: Roll out low-risk traffic first

Start with workflows where output can be reviewed or retried. Keep fixed-model paths for strict QA, sensitive actions, and features that depend on one model's specific behavior.

Common API errors and what to do

The Smart Router endpoint uses normal HTTP error handling:

Set explicit client timeouts. Avoid unlimited retries: they increase latency, duplicate work, and can turn a temporary failure into an expensive retry storm.

Common Smart Router mistakes

Assuming Smart Router always chooses the cheapest model

Smart Router is positioned around cost-quality control, not a guarantee that every request uses the lowest-priced model. Evaluate effective cost on your own workload.

Treating the routed model as stable

The same task may not always resolve to the same candidate model. Use a fixed model when repeatability or model identity is required.

Sending image or video generation requests to evolink/auto

The current Smart Router scope is supported text and agent requests. Use explicit model IDs and the relevant image or video endpoint for media generation.

Ignoring response.model

If you do not record the selected model, you lose the main signal needed to explain changes in quality, latency, and usage.

Publishing a hardcoded candidate-model list

Candidate models can change. Link to the current documentation and analyze the models returned by real requests instead.

Recommended production pattern

Use Smart Router as an observable default, not an invisible dependency:

1. Keep evolink/auto behind an application configuration flag.
2. Log the returned model and operational metrics.
3. Compare routed traffic with a fixed-model baseline.
4. Pin explicit models for strict QA or model-specific features.
5. Review routing results whenever your prompt distribution or product workflow changes.

FAQ

What endpoint should I use for EvoLink Smart Router?

What model ID enables Smart Router?

How can I see which model handled a request?

Is EvoLink Smart Router always cheaper than a fixed model?

No. It is designed to manage cost-quality tradeoffs across mixed workloads, but effective cost depends on the requests, routed models, output length, retries, and quality requirements.

Does Smart Router always return the same model for the same prompt?

Do not rely on that behavior. Candidate availability and routing policy can change. Use an explicit model ID when model identity or deterministic evaluation matters.

Can Smart Router route image and video generation?

The current product scope is supported text and agent requests. Use explicit image or video model IDs for media generation.

Can I use streaming?

When should I replace Smart Router with a fixed model?

Pin a fixed model when a workflow has a proven winner, requires model-specific features, needs strict regression testing, or follows a controlled approval process.

Next step