docs: add llamay to the OpenAI-compatible providers section #24227

Description

Please add a llamay page to content/providers/04-openai-compatible-providers/, next to LM Studio.

Disclosure: I work on llamay. This is a vendor request, so weigh it accordingly.

llamay (https://llamay.com) is a local inference server for GGUF models from Kosa8 Labs. It's free to use but not open source. Downloads: https://llamay.com/download/. llamay serve -m <model> exposes an OpenAI-compatible API on http://127.0.0.1:11435/v1. No provider package is needed: createOpenAICompatible with a baseURL is the full setup.

Vendor-controlled AI SDK instructions

• AI SDK guide: https://llamay.com/docs/developers/#ai-sdk

Setup facts

• Base URL: http://localhost:11435/v1. -addr changes it.
• Auth: none by default. With llamay serve -api-key K, it takes Authorization: Bearer K, which is apiKey in createOpenAICompatible.
• Model ids: Ollama-style names (qwen2.5:7b) or Hugging Face references (hf.co/bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M). GET /v1/models lists the models on disk.
• Embeddings: llamay serve -m <chat model> -embed <embedding model> serves /v1/embeddings.

import { createOpenAICompatible } from '@ai-sdk/openai-compatible';
import { generateText } from 'ai';

const llamay = createOpenAICompatible({
  name: 'llamay',
  baseURL: 'http://localhost:11435/v1',
});

const { text } = await generateText({
  model: llamay('qwen2.5:7b'),
  prompt: 'Write a vegetarian lasagna recipe for 4 people.',
});

Tested with ai 7.0.157 and @ai-sdk/openai-compatible 3.0.81

• generateText, streamText, a multi-step tool call with stopWhen, Output.object structured output, and embed, all against llamay serve -m qwen2.5:0.5b -embed all-minilm.

Also checked against the requests @ai-sdk/openai-compatible sends

• Chat completions, plain and streamed. With stream_options.include_usage, usage arrives on the final chunk.
• Tool calls, including tool_choice: "required" and named tool choice.
• response_format json_schema, for structured output.
• /v1/embeddings with encoding_format: "float".

Code of Conduct

• I agree to follow this project's Code of Conduct