AI Models (LLM)

Devora provides an OpenAI-compatible reverse proxy that lets you access multiple AI models from different providers through a single unified endpoint.

How It Works

When you send a chat request, Devora acts as a proxy between you and the upstream AI provider:

┌─────────┐     ┌──────────────────┐     ┌─────────────────┐
│  You    │────▶│  Devora Proxy    │────▶│  AI Provider    │
│ (Client)│     │  /api/v1/ai      │     │  (OpenAI, etc.) │
└─────────┘     └──────────────────┘     └─────────────────┘

Devora handles authentication, model validation, routing, and streaming — you just use the standard OpenAI API format.

Getting Started

1. Get an API Key

Before using the LLM API, you need a Devora API key:

• Dashboard: Go to API Key Management to create or copy your key.
• Docs: See the API Keys documentation for how to create, revoke, and manage keys.

Your API key must be included in every request as a Bearer token:

Authorization: Bearer devora_your_api_key

2. Choose a Model

Models are managed inside Devora by administrators. The list below is auto-fetched live from our database — if an admin adds a new model, it appears here automatically.

Where do models come from?
Admins add models via the AI Providers dashboard. Each model has an ID, name, provider (owned_by), and optional routing config (baseUrl). The table above pulls directly from /api/v1/ai/models.

3. Send a Request

Use the base URL below with any OpenAI-compatible client or SDK.

Base Configuration

[!TIP] When using the OpenAI SDK, set base_url to https://devora.my.id/api/v1/ai. The SDK automatically appends /chat/completions.

Authentication

All requests require an API Key in the Authorization header:

Authorization: Bearer devora_your_api_key

If your key is missing, expired, or invalid, the API returns:

{
  "error": {
    "message": "Invalid API Key",
    "type": "invalid_request_error",
    "code": "invalid_api_key"
  }
}

Model Identifiers

Use the exact id from the table above (for example, gpt-4o, gemini-1.5-pro, etc.) as the model parameter in your request body.

Private Models

Models with an email prefix (for example, [email protected]/gpt-4o) are private. They are only accessible by:

• The owner of that email address.
• Users with the ULTRA role.

Model Statuses

Models can be in different states which affect their availability:

Request Format

The chat endpoint accepts a standard OpenAI-compatible JSON body:

{
  "model": "gpt-4o",
  "messages": [
    { "role": "system", "content": "You are a helpful assistant." },
    { "role": "user", "content": "Hello!" }
  ],
  "stream": false,
  "temperature": 0.7,
  "max_tokens": 2048
}

Supported Parameters

Example Usage

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    api_key="devora_your_key",
    base_url="https://devora.my.id/api/v1/ai"
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
    stream=True
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="")

cURL

curl https://devora.my.id/api/v1/ai/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer devora_your_api_key" \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": false
  }'

JavaScript / TypeScript

const res = await fetch("https://devora.my.id/api/v1/ai/chat/completions", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    Authorization: "Bearer devora_your_api_key",
  },
  body: JSON.stringify({
    model: "gpt-4o",
    messages: [{ role: "user", content: "Hello!" }],
    stream: false,
  }),
});

const data = await res.json();
console.log(data.choices[0].message.content);

Streaming Responses

Set stream: true to receive tokens in real-time via Server-Sent Events (SSE). This is ideal for chat interfaces.

Headers returned:

Content-Type: text/event-stream
Cache-Control: no-cache
Connection: keep-alive

Error Handling

Devora returns standard HTTP status codes and OpenAI-compatible error objects:

Example Error Response

{
  "error": {
    "message": "Model 'gpt-4o' is currently suspended.",
    "type": "access_denied",
    "code": 403
  }
}

Related Resources

• API Keys documentation — How to create and manage your Devora API key.
• AI Providers dashboard — Where admins add, edit, and configure models.
• List Models endpoint — Public JSON endpoint returning all available models.
• Internal routing docs — How Devora routes requests to upstream providers.

[!NOTE] The model table above is rendered by a React component (<ModelTable />) that fetches /api/v1/ai/models on page load. As soon as an admin adds or removes a model in the dashboard, the table updates automatically — no manual doc edits required.