The Nevastack workspace
A place to start.
Get to know the workspace, organize your first project, and understand how the pieces fit together.
A project is the starting point for your work. It keeps related datasets, training jobs, deployments, and keys in one place.
- Sign in and open Projects in the console.
- Create a project with a name that describes the use case, such as
Customer support. - Choose a data sensitivity label and create the project.
Start with one use case. A focused project makes it easier to compare approaches and keep the work organized.
Open the training workspace to explore the data-upload flow. Start with documents or a dataset relevant to the question you want your application to answer.
Keep source material focused and up to date. Remove duplicates and check that you have permission to use the data before including it in a project.
The right approach depends on whether you need access to knowledge, a change in model behavior, or a model built for a particular task.
| Approach | A useful starting point when… |
|---|---|
| Retrieval / RAG | Your application needs answers from documents that change over time. |
| Fine-tuning / LoRA | You need a particular response format, style, or task-specific behavior. |
| Full training | You have a specialized objective, enough data, and a dedicated compute budget. |
An OpenAI-compatible endpoint accepts a model name and a list of messages. You can use an HTTP request or the OpenAI SDK to call it.
Integration example. This requires a separately provisioned inference endpoint. Creating a deployment record in the workspace does not start a model server.
Set NEVA_BASE_URL to your endpoint’s API base URL, including its /v1 path where applicable. Set NEVA_API_KEY to that endpoint’s key and replace your-model-name with an available model.
curl "$NEVA_BASE_URL/chat/completions" \
-H "Authorization: Bearer $NEVA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "your-model-name",
"messages": [
{"role": "user", "content": "Summarize our refund policy."}
],
"stream": false
}'Request fields
| Field | Type | Description |
|---|---|---|
model | string | The model identifier exposed by your endpoint. |
messages | array | Conversation messages with a role and content. |
stream | boolean | Use false for a complete response, or true for streaming when supported. |
Sign in to manage your workspace. The API keys page organizes key records by project and scope.
For inference requests, use the credential issued by the service running your model. Pass it in the authorization header:
Authorization: Bearer YOUR_API_KEYKeep inference credentials in your server’s environment. Browser code should call your own backend rather than expose a private API key.
Use Usage to review workspace activity and Billing to find plan and invoice information.
Request limits, token allowances, and charges depend on the inference service and plan you use. Confirm those details before connecting a production workload.
Explore plans and pricing