Skip to main content

Documentation Index

Fetch the complete documentation index at: https://gcore.com/docs/llms.txt

Use this file to discover all available pages before exploring further.

Gcore generates an HTTPS endpoint for each deployment that accepts OpenAI-compatible API requests. This article covers locating the endpoint and sending inference requests using the Chat Completions API.

​
Endpoint URL

The endpoint URL is available on the Overview tab of the deployment detail page under Everywhere Inference > Deployments.
Deployment overview showing endpoint URL
The URL follows this pattern: 
https://model-<deployment-name>-<project-id>-<account-id>.ai.gcore.dev/
Deployment names are alphanumeric without hyphens. A name containing hyphens causes a 400 error on creation.

​
Chat completions

The endpoint implements the OpenAI Chat Completions API at /v1/chat/completions. The model field must match the identifier returned by /v1/models — use the exact string, including the organization prefix (e.g., meta-llama/Llama-3.2-1B-Instruct). By default, API key authentication is disabled and the api_key / apiKey field is not validated — pass any non-empty string. When authentication is enabled on the deployment, include the key in the X-API-Key header on every request. See Inference deployment with API key authentication for details. 
curl -X POST "<ENDPOINT_URL>/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<MODEL_NAME>",
    "messages": [
      {"role": "user", "content": "Explain what a large language model is in two sentences."}
    ],
    "max_tokens": 200
  }'

​
Multi-turn conversation

Each request is stateless — the model has no memory of previous calls. To maintain context, pass the full conversation history in the messages array on every request. The system role sets the model’s overall behavior and is placed first in the array. It is optional but recommended for chat applications. As the conversation grows, the combined token count of all messages plus the generated reply must stay within the model’s context window. If the responses degrade or the request fails with a context length error, trim the oldest user/assistant exchanges from the history while keeping the system message. 
curl -X POST "<ENDPOINT_URL>/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<MODEL_NAME>",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "What is the capital of France?"},
      {"role": "assistant", "content": "The capital of France is Paris."},
      {"role": "user", "content": "What is its population?"}
    ],
    "max_tokens": 100
  }'

​
Streaming responses

With stream: true, the server returns the response as server-sent events (SSE). Each event is formatted as data: {...} followed by a blank line; the sequence ends with data: [DONE]. The generated text arrives in choices[0].delta.content — this field is null on the first and last chunks. Streaming reduces time-to-first-token latency and suits interactive chat UIs. 
curl -X POST "<ENDPOINT_URL>/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<MODEL_NAME>",
    "messages": [{"role": "user", "content": "Write a short poem about cloud computing."}],
    "max_tokens": 150,
    "stream": true
  }'

​
Available models

The /v1/models endpoint returns the model identifiers accepted by the deployment: 
curl "<ENDPOINT_URL>/v1/models"
The id field in the response is the value to pass as model in chat completion requests.

​
Key parameters

ParameterTypeDescription
modelstringModel identifier. Use the id value from the /v1/models response.
messagesarrayConversation history. Each item has role (system, user, or assistant) and content.
max_tokensintegerMaximum number of tokens to generate.
temperaturefloatRandomness of output. Range 0–2. Lower values are more deterministic. Default: 1.
top_pfloatNucleus sampling threshold. Range 0–1. Default: 1.
streambooleanIf true, responses are sent token by token as server-sent events. Default: false.
stopstring or arrayOne or more sequences where generation stops.

​
Endpoint authentication

By default, the endpoint is publicly accessible. To restrict access, enable API key authentication as described in Inference deployment with API key authentication.