LlaDash API Documentation

The LlaDash API is a keyless, high-performance routing API. It dynamically splits tasks between Llama 3.1 8B for fast operations and ChatGPT OSS 120B for deep reasoning.

[Streaming Behavior] When using ?stream=true in plain text mode, internal <think> tags and their contents are automatically stripped from the stream, delivering only the clean final text. JSON/XML streaming preserves reasoning within dedicated fields.

Architecture Overview

Two robust models dynamically handle processing depending on the query tag:
1. Llama 3.1 8B (Fast): Performs safety moderation, route tag selection (clock/calc/search/think), quick answers, and localized title generation.
2. ChatGPT OSS 120B (Pro): Formulates high-level step-by-step reasoning, handles complex inquiries, and summarizes deep web search results.

Text Generation Endpoint

Send a prompt to receive AI inferences. Supports both GET and POST requests.

GET POST /ai/{prompt}

Query Parameters

Name	Type	Description
`prompt`	String	Max length is 2000 characters. Exceeding this returns 400 Bad Request.
`json`	Boolean	Returns results in JSON, containing structured thinking and title fields if active.
`xml`	Boolean	Returns results in XML format.
`stream`	Boolean	Streams the response in real-time. Emits purely clean raw response text directly without SSE prefixes in text mode.
`title`	Boolean	Generates a localized title (under 15 chars) for every single query automatically.

Example Request (JSON + Title)

const url = 'https://api.ndnx.workers.dev/ai/What is AI?&json=true&title=true';
const response = await fetch(url, {
  headers: { 'User-Agent': 'MyAwesomeApp/1.0' }
});
const data = await response.json();
console.log(data);

Example Response

{
  "status": "success",
  "model": "openai/gpt-oss-120b",
  "thinking": "The user is asking for a definition of AI...",
  "title": "AIの定義",
  "answer": "人工知能（AI）とは、人間の知的振る舞いをコンピュータ上で再現する..."
}

Rate Limits

System-wide concurrent queue limits and model-specific quotas.

GET /ai/rate

Rate Limits

System Queue Limit: Max 20 concurrent requests system-wide (Returns HTTP 503 if busy).
Single IP Lock: Only 1 active request per client allowed at a time (Returns HTTP 429 on multi-send).
Llama 3.1 8B (Fast Model / Router): 200 Requests Per Minute / 10,000 requests per day limit.
ChatGPT OSS 120B (Pro Model / Generator): 30 Requests Per Minute / 1,000 requests per day limit.

Error Responses

400 Bad Request: Prompt too long or invalid request.
403 Forbidden: Bot detected, missing User-Agent, or security block.
429 Too Many Requests: Global queue limit (20/min) exceeded.
500 Internal Server Error: Backend API error or timeout.