@dartix-software-solutions/ai-gate weekly downloads statistics

@dartix-software-solutions/ai-gate

Official TypeScript / JavaScript client for AI Gate — one API key for chat and image models across many providers, with latency-tuned auto routing and automatic failover.

Keywords: AI · LLM · SDK · OpenAI · Gateway · OpenRouter · NVIDIA · TypeScript · Gemini · Groq · Anthropic · Cerebras · Together · DeepSeek · Hugging Face

---

What this package does

AI Gate sits between your app and AI providers. Instead of wiring each vendor SDK yourself, you call this client once.

Your app  →  AI Gate (this SDK)  →  Gemini / NVIDIA / OpenRouter / Groq / …

Capability	What you get
Chat generation	Send a prompt (and optional context) and get text back
Streaming chat	Token-by-token responses over SSE
Image generation	Text-to-image; returns a URL or data URI
Model catalog	List available chat / image models by free or premium tier
Latency-tuned auto routing	Picks a healthy, low-latency free model when you omit model
Failover	Retries other models/providers in the same pool if one fails
Dual-key tiers	Free via platform keys; premium via keys you save on the project
Usage metadata	Provider used, tokens, latency, and full attempt history
Daily project quotas	30 requests + 30,000 API tokens / UTC day (see Usage limits)
Internet + reasoning	Optional internet and reasoningLevel on generate / stream

Dual-key model (how keys work)

Tier	When	Keys needed
free (default)	You don’t ask for premium	AI Gate project key only — free models use platform provider keys
premium	tier: "premium"	Project key + your provider API key on the project (any provider: OpenAI, OpenRouter, Hugging Face, Groq, …)
auto	Mix free + available premium	Free via platform keys; premium only for providers where you saved a key

Chat and image never mix. Failover stays inside the same modality + tier pool.

Supported providers

Free (platform keys)	Premium (your project keys)
Gemini, NVIDIA NIM, OpenRouter, Pollinations, Groq, Cerebras, Together AI, DeepSeek, Hugging Face, Eden AI, ShareAI	Same providers plus OpenAI, Anthropic — add any provider key on the project to unlock that provider’s premium models

---

What’s included

This npm package ships:

Export	Description
AIClient	Main client class
AIClientOptions	Constructor options (apiKey)
generate()	Non-streaming chat → POST /v1/generate
stream()	Streaming chat → POST /v1/generate/stream
image()	Image generation → POST /v1/images/generate
listModels()	Catalog → GET /v1/models
TypeScript types	GenerateRequest, GenerateResponse, ImageGenerateRequest, ImageGenerateResponse, AIChunk, CatalogModel, Tier, Modality, …

No extra runtime dependencies (uses Node’s built-in fetch). Types are bundled — you do not need a separate @ai-gate/types install.

---

Requirements

• Node.js 18+ (also works with Bun / Deno where fetch is available)
• An AI Gate project API key (ai_live_…) from aigate.dartix.live

---

Install

npm install @dartix-software-solutions/ai-gate

pnpm add @dartix-software-solutions/ai-gate

yarn add @dartix-software-solutions/ai-gate

---

How to use

1. Get an API key

1. Open aigate.dartix.live and sign in
2. Create a project
3. Copy the project key (ai_live_…)
4. Store it in env on your server / BFF — never ship it in browser-exposed client code

AI_GATE_API_KEY=ai_live_xxxxxxxx

2. Create a client

import { AIClient } from "@dartix-software-solutions/ai-gate";

const ai = new AIClient({
  apiKey: process.env.AI_GATE_API_KEY!,
});

Option	Required	Description
apiKey	Yes	Project API key from the dashboard

3. Chat (generate)

Defaults: tier: "free", chat modality, auto model (fastest healthy free chat model via latency-tuned routing).

const result = await ai.generate({
  prompt: "Write a short product description",
  context: {
    productName: "Nike Air Max",
    category: "Running",
  },
});

console.log(result.content);
console.log(result.provider); // { name, model, tier }
console.log(result.usage);    // tokens
console.log(result.latency);  // ms
console.log(result.attempts); // failover trail (includes latencyMs per try)

Optional fields:

Field	Default	Notes
tier	"free"	"free" | "premium" | "auto"
provider / model	unset / auto	Pin first choice; failover stays in pool
maxTokens	server default	Cap output (max 4000)
internet	false	When true, free DuckDuckGo results may be injected for live-info questions
reasoningLevel	"auto"	"auto" | "low" | "medium" | "high"

// Pin provider or model (failover still applies within the pool)
await ai.generate({ prompt: "…", provider: "openrouter" });
await ai.generate({ prompt: "…", provider: "nvidia" });
await ai.generate({ prompt: "…", provider: "groq" });
await ai.generate({ prompt: "…", model: "gemini-flash" });…