Moondream2 Visual Query API Reference

AUTHENTICATION

Bearer Token Authentication

All requests to the LumozAI API require authentication using your secret API key. Pass your key in the standard Authorization HTTP header as a Bearer token.

cURL Authorization Header Quickstart

curl -X POST https://lumozai.com/api/v4/run \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Accept: application/json" \
  -d "model_id=moondream2-visual-query" \
  -d "prompt=A futuristic cityscape illuminated by neon bioluminescence"

PHP cURL Request

<?php
$apiKey = 'YOUR_API_KEY';

$curl = curl_init();
curl_setopt_array($curl, [
    CURLOPT_URL => 'https://lumozai.com/api/v4/run',
    CURLOPT_RETURNTRANSFER => true,
    CURLOPT_POST => true,
    CURLOPT_POSTFIELDS => [
        'model_id' => 'moondream2-visual-query',
        'prompt' => 'A futuristic cityscape illuminated by neon bioluminescence'
    ],
    CURLOPT_HTTPHEADER => [
        'Accept: application/json',
        'Authorization: Bearer ' . $apiKey
    ]
]);

$response = curl_exec($curl);
curl_close($curl);

$result = json_decode($response, true);
print_r($result);

POST /api/v4/run

Execute Moondream2 Visual Query Inference

Submit a synchronous or asynchronous generation job for Moondream2 Visual Query (moondream2-visual-query). Fast-running image models return the generated asset directly; longer inference pipelines return a request_id for asynchronous polling.

Request Parameters for Moondream2 Visual Query

4 total fields

Parameter	Type	Required	Default	Description / Allowed Values
`model_id`	string	Required	`moondream2-visual-query`	Unique model identifier. Must be set to `moondream2-visual-query`.
`image_url` Image	string (URL)	Required	—
`query` Query	string	Required	—
`webhook`	string (URL)	Optional	—	Optional callback URL to receive an HTTP POST when async generation completes.

Model-Specific Integration Code

Copy-paste production snippets configured with moondream2-visual-query and tailored parameter inputs.

cURL POST /api/v4/run (moondream2-visual-query)

curl -X POST https://lumozai.com/api/v4/run \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Accept: application/json" \
  -d "model_id=moondream2-visual-query" \
  -d "image_url=https://lumozai.com/samples/sample_input.jpg" \
  -d "query=sample_value"

Python requests.post (moondream2-visual-query)

import requests

url = "https://lumozai.com/api/v4/run"
headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Accept": "application/json"
}
payload = {
    "model_id": "moondream2-visual-query",
    "image_url": "https:\/\/lumozai.com\/samples\/sample_input.jpg",
    "query": "sample_value"
}

response = requests.post(url, headers=headers, json=payload)
print(response.json())

PHP cURL POST (moondream2-visual-query)

<?php
$apiKey = 'YOUR_API_KEY';

$curl = curl_init();
curl_setopt_array($curl, [
    CURLOPT_URL => 'https://lumozai.com/api/v4/run',
    CURLOPT_RETURNTRANSFER => true,
    CURLOPT_POST => true,
    CURLOPT_POSTFIELDS => [
        'model_id' => 'moondream2-visual-query',
        'image_url' => 'https://lumozai.com/samples/sample_input.jpg',
        'query' => 'sample_value'
    ],
    CURLOPT_HTTPHEADER => [
        'Accept: application/json',
        'Authorization: Bearer ' . $apiKey
    ]
]);

$response = curl_exec($curl);
curl_close($curl);

$result = json_decode($response, true);
print_r($result);

JavaScript fetch async/await (moondream2-visual-query)

const response = await fetch('https://lumozai.com/api/v4/run', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json',
    'Accept': 'application/json'
  },
  body: JSON.stringify({
    "model_id": "moondream2-visual-query",
    "image_url": "https:\/\/lumozai.com\/samples\/sample_input.jpg",
    "query": "sample_value"
  })
});

const data = await response.json();
console.log(data);

Response Structure (200 OK)

JSON Output

{
  "success": true,
  "data": {
    "request_id": 982348712938172,
    "output": "https://lumozai.com/storage/outputs/result_moondream2-visual-query_9823.png",
    "output_type": "image"
  }
}

POST /api/v4/check

Poll Asynchronous Task Status

Poll the current execution state of an asynchronous job using the returned request_id.

Request Parameters

Parameter	Type	Required	Description
`request_id`	integer	Required	The unique job ID returned when calling `/api/v4/run`.

JSON Completed Status (200 OK)

{
  "success": true,
  "data": {
    "request_id": 982348712938172,
    "status": "completed",
    "output": "https://lumozai.com/storage/outputs/video_8172.mp4",
    "output_type": "video"
  }
}

JSON In-Progress Status (200 OK)

{
  "success": true,
  "data": {
    "request_id": 982348712938172,
    "status": "processing",
    "progress": 68
  }
}

ERROR HANDLING

Standard Error Codes

The API uses conventional HTTP response codes to indicate success or failure. In general, 2xx indicates success, 4xx indicates client input issues, and 5xx indicates server errors.

Status	Error Code	Description
401	`authentication_failed`	Missing or invalid Authorization Bearer header.
403	`permission_denied`	Insufficient credits or account tier permissions.
404	`model_not_found`	The specified `model_id` does not exist or is deprecated.
422	`validation_error`	Missing required parameters or invalid parameter types.
429	`rate_limit_exceeded`	Request rate quota exceeded for your tier.
500	`internal_server_error`	Downstream GPU compute error or worker timeout.

JSON Validation Error Example

{
  "success": false,
  "error": {
    "code": "validation_error",
    "message": "The model_id field is required.",
    "details": {
      "model_id": ["The model_id field is required."]
    }
  }
}

RATE LIMITS

Quota & Throughput Limits

Rate limits are enforced per API key to guarantee fair access and consistent low latency across GPU workers. Every API response includes informative headers:

Header	Description
`X-RateLimit-Limit`	Maximum number of requests permitted per rolling hour.
`X-RateLimit-Remaining`	Number of remaining requests in the active window.
`X-RateLimit-Reset`	Unix epoch timestamp when the current window refreshes.

Handling 429 Status Codes: When rate-limited, wait for the number of seconds until X-RateLimit-Reset or implement an exponential backoff retry mechanism. Need custom enterprise rate limits? Contact our team.

NEW API /api/v1/memory/*

Serverless Vector Memory & Semantic Search

Store, index, and retrieve vector embeddings with sub-50ms latency. Unlike traditional vector databases that charge $25 to $70 every month just to keep a cluster active, Lumoz offers zero base fees: 5,000 vector memory slots and 10,000 semantic queries per month are included free with any active balance reload.

Core Endpoints

Method	Endpoint	Description
GET	`/api/v1/memory/usage`	Inspect active tier, vector storage meter, and monthly search counters.
POST	`/api/v1/memory/projects`	Create an isolated project/namespace (e.g. `support-bot`).
POST	`/api/v1/memory/store`	Embed and index text documents or knowledge chunks.
POST	`/api/v1/memory/search`	Cosine similarity search with hybrid keyword weighting.
DELETE	`/api/v1/memory/{id}`	Delete an indexed memory item.

Code Examples

cURL 1. Store Knowledge & 2. Semantic Search

# 1. Ingest text into an isolated namespace
curl -X POST "https://lumozai.com/api/v1/memory/store" \
  -H "Authorization: Bearer YOUR_LUMOZ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "project": "support-bot",
    "text": "Refunds are processed within 48 hours to original payment method.",
    "metadata": {"category": "billing"}
  }'

# 2. Query vector memory via cosine similarity
curl -X POST "https://lumozai.com/api/v1/memory/search" \
  -H "Authorization: Bearer YOUR_LUMOZ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "project": "support-bot",
    "query": "How long do refunds take?",
    "top_k": 3
  }'

Python requests (Vector Store & Search)

import requests

headers = {
    "Authorization": "Bearer YOUR_LUMOZ_API_KEY",
    "Content-Type": "application/json"
}

# Store Knowledge
requests.post("https://lumozai.com/api/v1/memory/store", headers=headers, json={
    "project": "support-bot",
    "text": "Enterprise SLA guarantees 99.99% uptime with dedicated GPU routing.",
    "metadata": {"tier": "enterprise"}
})

# Semantic Search
res = requests.post("https://lumozai.com/api/v1/memory/search", headers=headers, json={
    "project": "support-bot",
    "query": "What is the uptime SLA?",
    "top_k": 2
})
print("Top Matches:", res.json()["results"])

Zero-Code RAG with Lumoz LLM

Pass "memory": {"project": "..."} directly to /api/v1/chat/completions. Lumoz automatically pulls relevant memories into the prompt context without requiring any custom client-side vector code:

Node.js Zero-Code RAG Chat Completion

const response = await fetch("https://lumozai.com/api/v1/chat/completions", {
  method: "POST",
  headers: {
    "Authorization": "Bearer YOUR_LUMOZ_API_KEY",
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    model: "lumoz-llm-1.0",
    messages: [
      { role: "user", content: "What is your refund policy window?" }
    ],
    memory: {
      project: "support-bot",
      top_k: 3
    }
  })
});

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