The best open-source tools for translators, in one place, ready to use.
The open-source world has excellent tools for the jobs around translation — reading PDFs properly, running a language model on your own machine, transcribing speech. The Integrations screen brings them into the tool you translate in, and a configuration template for each means you choose what you want rather than working out how to wire it up.
Pick what you want to run — an AI model server on this PC or your network, an open model on a GPU you rent, your own cloud endpoint, a document reader for PDFs — and the template fills in the address, the protocol and the paths. You add what only you know, and test it before you save it.
Point it at an AI model server you already run — Ollama, LM Studio, llama.cpp — and the connection test lists the AI models on it, so you pick one rather than type its name. Or rent a GPU by the hour and run an open model on it, from Gemma and Qwen to ApertusThe open multilingual model from the Swiss AI Initiative (EPFL, ETH Zurich, CSCS), Apache 2.0 — offered on a rented GPU, including a provider with a Swiss zone., or any AI model you name. Nothing about your work leaves the machines you chose.
A PDF is prepared before it is translated, and the reader is yours to choose: Docling on your own machine, the official Docling server, or pdf4llm. The result is a file the ordinary import reads.
Point it at a commercial provider, at a machine on your own network, at a GPU you rent by the hour, or at an endpoint in your own cloud account. Endpoints you run yourself do not need an API key.
The list is a published catalogue refreshed when the application starts, so a new model appears without a new version of AV. Models you have starred are kept separately and survive the refresh.
The first run walks through storage, options and keys, and lets you test each key as you add it. The provider list it offers comes from the application itself, so it is always current. Any key you skip, or one for a provider you take up later, can be added in LLM Configuration whenever you need it. Every key is kept in an encrypted file on your PC, and the key to that file is locked by Windows to your own Windows sign-in, the protection Windows itself uses for saved passwords: only you, signed in on this PC, can open it.
Every integration that receives document content carries a label you set: Internal, inside your network; External · private, with a provider you contract; or External, a third party. The application proposes one from the endpoint's shape and you confirm it — and if a project that protects nothing is about to send to an External endpoint, it tells you so, by name.
