AI-Studio

MindWork AI Studio is a free desktop application that lets you interact with large language models (LLMs) from a variety of providers. It runs on macOS, Windows and Linux, and you can use it for commercial projects without paying a license fee, although the developers note that donations are appreciated.

The core of the app is a simple prompt‑response interface. You start by selecting an LLM, either a cloud service you have credentials for or a model you have installed locally. The input is plain text entered into a textbox; the output appears in the same window as a generated text block. Because the client can talk to any provider that offers an API, you can switch between models without leaving the application, which is useful for comparing performance or cost. The tool is aimed at users who need a consistent workspace for daily LLM work, such as writers, researchers, or developers who prototype prompts.

A distinctive capability is Retrieval‑Augmented Generation (RAG). If you want the model to reference external knowledge, you can attach a data source that implements the External Retrieval Interface (ERI). An ERI‑compatible source acts as a server that stores its own vector database and exposes an API. When you submit a prompt, AI Studio sends the query to the ERI server, receives the most relevant vectors, and includes them in the prompt sent to the LLM. In practice, you would configure the ERI endpoint URL and any required authentication, then select it as a retrieval source. The result is a response that draws on up‑to‑date information without the data ever leaving its original host, which can reduce hallucinations on factual queries.

For organizations that keep data on‑premises, AI Studio also offers a beta local RAG feature. You point the app at a folder or a set of files; the program creates temporary embeddings using an internal method and stores them locally. When you ask a question, the app searches those embeddings and injects the most relevant snippets into the prompt. Typical inputs are plain‑text documents, markdown files, or code snippets; outputs remain plain text generated by the selected LLM.