Make your first request in minutes.

1. 1

   Create an inference key

   Sign in, open API keys, and create an nvai_… key. The full secret is shown once, so store it securely.

2. 2

   Set your environment variable

   Keep the key outside source control.

   terminalCopy

   export NEVIRI_AI_KEY="nvai_your_api_key_here"

3. 3

   Send a chat completion

   Use the base URL https://api.ai.neviri.com/v1 with a model available in your environment.

Use raw HTTP or the official OpenAI SDK. Install the SDK first with pip install openai or npm install openai.

curl https://api.ai.neviri.com/v1/chat/completions \
  -H "Authorization: Bearer $NEVIRI_AI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "neviri/flagship-chat",
    "messages": [
      { "role": "user", "content": "Explain edge computing in one sentence." }
    ]
  }'

Expected response

Read generated text from choices[0].message.content. Token counts are returned in usage.

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "model": "neviri/flagship-chat",
  "choices": [{
    "index": 0,
    "message": { "role": "assistant", "content": "Edge computing..." },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 14,
    "completion_tokens": 18,
    "total_tokens": 32
  }
}

Model availability is deployment-specific. Query the catalogue before relying on a slug; the endpoint returns configured model ids, context windows, modalities, and reference pricing metadata.

curl https://api.ai.neviri.com/v1/models

Set stream: true to receive incremental OpenAI-compatible chunks. Raw HTTP clients receive Server-Sent Events ending with data: [DONE].

const stream = await client.chat.completions.create({
  model: "neviri/flagship-chat",
  messages: [{ role: "user", content: "Write a two-line poem." }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

Errors include a machine-readable code and a message. Do not log API keys or complete request payloads while diagnosing failures.

Neviri currently exposes OpenAI-compatible chat completions and model discovery. Chat requests accept messages, streaming, sampling controls, stop sequences, penalties, seeds, JSON response format, and tool definitions. Upstream model support can vary.

These items are planned, not currently available commitments. They will move into the main documentation only after their behavior is implemented and verified.

• Embeddings API and integration examples
• Published rate limits, response headers, and 429 guidance
• Verified framework recipes for LangChain, LlamaIndex, AutoGen, and CrewAI
• Interactive request builder and downloadable OpenAPI specification
• Dedicated client and agent SDKs
• AI-assistant access through an MCP documentation server
• Provider controls, automatic fallbacks, presets, and routing policies