Best practices

These recommendations help you keep cost, speed and reliability under control when working with the codexkey API.

Choose the right model

Match the model to the task complexity:

For simple tasks a smaller model cuts both cost and latency noticeably.

Use prompt caching

When the same system prompt goes out with many requests, turn on caching: it makes input tokens cheaper.

Mark the system message block with the cache_control field:

{
  "system": [
    {
      "type": "text",
      "text": "You are a helpful coding assistant...",
      "cache_control": { "type": "ephemeral" }
    }
  ]
}

Set appropriate max_tokens

Set max_tokens to the smallest value that works:

• Short answers: 256–512
• Code generation: 2048–4096
• Long-form content: 4096–8192

Lower values cut cost and sometimes reduce latency.

Use streaming for real-time responses

Turn streaming on: tokens arrive as they are generated, and the answer feels faster to the user:

curl https://api.codexkey.online/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: sk_cdx_YOUR_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 1024,
    "stream": true,
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Use separate keys per environment

Create one key per environment, so usage does not mix and debugging gets easier:

• Production: a named key with a suitable channel and expiry
• Development: a separate key, so test spend is counted on its own
• CI/CD: a dedicated key for automated tests

That makes it simpler to watch spending, rotate keys and revoke access for each environment.

Monitor usage and set alerts

In the analytics dashboard, watch:

• Token consumption per key and model
• Cost trends over time
• Request counts by key
• Unexpected usage spikes

Turn on low-balance alerts in Settings so you hear about it before the funds run out.