One unified, provider-agnostic API for 10 AI model providers — OpenAI, Anthropic, Gemini, Groq, and more. Write ai.chat() once and swap providers with a config change, never a code change.

@radonsdk/ai gives you one typed API — chat, stream, embed — that every one of 10 AI providers implements. Provider-specific quirks (auth schemes, wildly different tool-calling formats, SSE framing, JSON-mode dialects) are absorbed inside each adapter and never leak into your code.

ask.ts

import { RadonAI } from "@radonsdk/ai";

const ai = new RadonAI({
  providers: { openai: {} },      // credentials from RADON_OPENAI_API_KEY
  defaultProvider: "openai",
});

const res = await ai.chat({
  messages: [{ role: "user", content: "Say hello in one word." }],
});

res.content;        // "Hello"
res.usage;          // { promptTokens, completionTokens, totalTokens }
res.finishReason;   // "stop" | "length" | "tool_calls" | ...

Switching to Anthropic, Gemini, or Groq later is a one-line config change — the chat() call above never changes.

Own your keys, no lock-in

Radon runs in your backend against your own provider keys. There's no hosted gateway, no per-token fee from Radon, and the SDK never stores your secrets — each adapter reads them from RADON_<PROVIDER>_API_KEY at call time.

What's in the box

Quickstart

Send your first message in under five minutes.

Core concepts

Providers, the one message/result shape, models, streaming, and the escape hatch.

Chat

Messages, roles, sampling parameters, and the normalized result.

Streaming

Token-by-token deltas, normalized identically across providers.

Tool calling

Define one tool schema; it works on every provider. The point of the SDK.

Providers

All 10 providers, their env vars, default models, and quirks.

API reference

Every method, option, and return type.

Errors

Every typed error, its code, and what causes it.

Free vs. Pro

Three providers are free and need no license: OpenAI, Anthropic, Groq — with full chat, streaming, and normalized tool-calling. Tool-calling is core, never an upsell.

The other seven providers — Google (Gemini), Mistral, DeepSeek, OpenRouter, xAI (Grok), Together AI, Ollama — are Pro Pro, as are a handful of features on any provider: embeddings, vision / image input, structured output / JSON mode, and fallback chains. All of these unlock with a single Radon Pro license, verified once in await ai.init().

Gating is a licensing concern only — every adapter, free or Pro, is built to the same completeness bar: real API calls, real streaming, real tool-calling.

Radon AI