Chat Completions - Valkyrie

Valkyrie serves deployed models through the OpenAI Chat Completions contract.

Endpoint

POST https://valkyrie-back.azumo.com/{deployment_slug}/v1/chat/completions

With an alias:

POST https://valkyrie-back.azumo.com/{account_slug}/{alias}/v1/chat/completions

An alias can point at a deployment or an inference pool. A pool alias sends each request to a pool member, needs an account-wide API key, and uses the pool’s model whatever model you send. Send X-Session-Affinity to keep a session on one member. See Point an alias at an inference pool.

Request

Send the standard OpenAI Chat Completions body. Common fields:

Field	Type	Notes
`model`	string	Your deployment slug or alias
`messages`	array	Conversation, each `{ "role", "content" }`
`temperature`	number	Sampling temperature (optional)
`max_tokens`	integer	Maximum tokens to generate (optional)
`stream`	boolean	Stream tokens as server-sent events (optional)

curl https://valkyrie-back.azumo.com/{deployment_slug}/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $VALKYRIE_API_KEY" \
  -d '{
    "model": "{deployment_slug}",
    "messages": [
      {"role": "system", "content": "You are concise."},
      {"role": "user", "content": "What is Valkyrie?"}
    ],
    "temperature": 0.7,
    "max_tokens": 256
  }'

Response

The response follows the OpenAI shape:

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "model": "{deployment_slug}",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "..." },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 0,
    "total_tokens": 0
  }
}

Streaming

Set "stream": true to receive incremental chat.completion.chunk events as server-sent events, terminated by a [DONE] sentinel, identical to the OpenAI streaming format. The OpenAI SDKs expose this through their streaming helpers.

Supported sampling parameters and context length depend on the deployed model and its serving engine. If a deployment is stopped, requests fail until it is resumed.

Inference API

​Endpoint

​Request

​Response

​Streaming

Endpoint

Request

Response

Streaming