@kyma-api/ai-sdk

Kyma API provider for the Vercel AI SDK.

Access 20+ state-of-the-art open-source models (DeepSeek, Qwen, Llama, Gemma, Gemini, Kimi, MiniMax and more) through a single OpenAI-compatible endpoint.

Installation

npm install @kyma-api/ai-sdk ai

Setup

Get a free API key at kymaapi.com — includes free credits on signup.

export KYMA_API_KEY=ky-...

Usage

import { kyma } from "@kyma-api/ai-sdk";
import { generateText } from "ai";

const { text } = await generateText({
  model: kyma("deepseek-v3"),
  prompt: "Explain quantum computing in one paragraph.",
});

With streaming

import { kyma } from "@kyma-api/ai-sdk";
import { streamText } from "ai";

const { textStream } = await streamText({
  model: kyma("llama-3.3-70b"),
  messages: [{ role: "user", content: "Write a haiku about the ocean." }],
});

for await (const chunk of textStream) {
  process.stdout.write(chunk);
}

Custom provider instance

import { createKyma } from "@kyma-api/ai-sdk";

const kyma = createKyma({
  apiKey: "ky-your-key-here",
});

Smart aliases

Instead of memorizing model IDs, use built-in aliases:

kyma("best")          // → qwen-3.6-plus (highest quality)
kyma("fast")          // → gpt-oss-120b (fastest)
kyma("code")          // → kimi-k2.5 (best for coding agents)
kyma("reasoning")     // → deepseek-r1 (deep reasoning)
kyma("long-context")  // → gemini-2.5-flash (1M context)
kyma("vision")        // → gemma-4-31b (multimodal)

Available Models

Tier 1 — Highest Quality

| Model ID | Description | |---|---| | qwen-3.6-plus | Alibaba's flagship. #1 on Kyma. | | deepseek-v3 | GPT-5 class. Best value. | | deepseek-r1 | Top reasoning model. 96% cheaper than o1. | | kimi-k2.5 | Multimodal agentic. 262K context. | | gemma-4-31b | Google's best open model. Multimodal. | | qwen-3-32b | Top coding. Ultra-fast inference. | | llama-3.3-70b | Most popular open model. | | minimax-m2.5 | SWE-bench 80.2%. Top agentic coding. |

Tier 2 — High Quality

| Model ID | Description | |---|---| | kimi-k2 | Fast agentic coding on Groq. | | minimax-m2.7 | Agentic productivity. | | gpt-oss-120b | OpenAI open source 120B. | | qwen-3-coder | Purpose-built for code. | | gemma-4-26b-moe | Efficient MoE. Only 4B active params. | | nemotron-3-super | NVIDIA 120B MoE. | | gemini-2.5-flash | Google fast model. 1M context. | | gemini-3-flash | Newest Gemini. 1M context. |

Tier 3 — Fast & Cheap

| Model ID | Description | |---|---| | llama-4-scout | MoE. 512K context. | | gemini-2.5-flash-lite | Lightest Gemini. 1M context. | | step-3.5-flash | StepFun reasoning. | | glm-4.5-air | Zhipu AI. Strong reasoning. |

Pricing

All models available with free credits on signup. Pay-per-token after that — no subscriptions, no seats.

See full pricing at kymaapi.com.

Links

• Kyma API
• Documentation
• Model list
• GitHub

License

MIT