Adding another LLM
You connect another LLM from the same Settings screen as a private machine translation engine. Once it is ticked for the project, its suggestion appears next to the other engines. You can compare them on the open segment and keep the wording that fits the job. Choosing it as the assistant is a second step, on that model’s page.
On the project creation page, open Settings with the Settings button. In a project you are already translating, the same screen opens from Project Settings in the editor header. The MT and LLM tab is the same in both places.
The listed models fill in the server address for you. The list includes Claude, Gemini, Groq, Grok, Mistral, Perplexity, OpenRouter, DeepSeek, Qwen, and ChatGPT if you prefer to use your own OpenAI key. The form links to the page where that provider gives you a key, for example Get a Claude API key.
The Engine name is filled in for a listed model. Keep it to 15 characters: that name is what you see in the suggestion panel and in the Provider menu. Description is optional. Leave it blank to show the model name in the engine list.
Get a subscription, or an API key, on the provider site first. Then add the engine in Wordscope.
To connect an LLM from an open project:
1️⃣ Open a project.
2️⃣ Click Project Settings.
3️⃣ Click the MT and LLM tab.
4️⃣ Click Add a private engine.
5️⃣ Click LLM.
6️⃣ Click the provider card. For example, Claude or Gemini.
7️⃣ In API key, paste the key from that provider.
8️⃣ Click Load models.
9️⃣ Click the model you want.
Confirm stays grey until Engine name, API key, and Model are filled. For a listed provider, Base URL is already filled and locked. If the service has no model list, type the model id in Model (for example claude-sonnet-4-6 or perplexity/sonar).
Perplexity needs more than a basic account. Wordscope asks for the open segment and the next two at the same time, and a basic Perplexity account allows only 1 call per second. You need at least $50 of API credits so this engine does not slow Wordscope down (tier 1: 3 calls per second).
DeepSeek and Qwen are Chinese LLMs. You use them under your own responsibility. grok-4.7, grok-4.6, and grok-4.5 are slower for translation. They default to high reasoning and cannot turn reasoning off.
If you share or outsource the project, the engines you attach are shared with the people on that project. If you are working on a project someone else created, an LLM you add stays private to you. You cannot change the owner’s engine list.
Check the key before you save
The form can send one sentence and one glossary pair before you click Confirm. Wordscope uses the same request it uses for a suggestion. This is worth doing on a medical or technical job. The translation should include the target glossary term. That is how you know the model will follow your glossary in a project.
To test the key:
1️⃣ In Source sentence, keep the example or type a sentence from the job.
2️⃣ In Source, type the source term.
3️⃣ In Target, type the term you want in the translation.
4️⃣ Open Target language and choose the language.
5️⃣ Click Test.
6️⃣ Read the translation. It should include the target glossary term.
Saving the engine is the last click, whether you ran the test or skipped it.
To save the engine:
1️⃣ Click Confirm.
Click Close to return to the editor. Use in this project is ticked, so the new engine’s suggestion appears with the others. Untick it on a project where you do not want that model. You can add as many engines as you want. Asking one more engine does not make the suggestion panel wait longer.
Connect your own or a local model
Use OpenAI-compatible LLM when the model is not in the list: an enterprise server, or a model running on a machine in your office. Wordscope calls your base URL plus /chat/completions, with your key and your model. A URL that already ends in /chat/completions is accepted too.
To connect that server from an open project:
1️⃣ Open a project.
2️⃣ Click Project Settings.
3️⃣ Click the MT and LLM tab.
4️⃣ Click Add a private engine.
5️⃣ Click LLM.
6️⃣ Click OpenAI-compatible LLM.
7️⃣ In Engine name, type a short name (15 characters maximum).
8️⃣ In Base URL, paste the address of your server.
9️⃣ In API key, paste the key.
10️⃣ In Model, type the model id.
If the server returns a model list, click Load models and click the model. That fills Model for you.
To save this engine:
1️⃣ Click Confirm.
A provider that is not listed, and that does not use this calling convention, cannot be added from this screen. Write to support@wordscope.com.
Choose it as the assistant
Connecting the engine adds it to the suggestion panel. ChatGPT stays the assistant until you pick the new engine in Provider. Each model has its own page for that step. Your custom prompts work on the new provider with no change.
Wordscope is an all-in-one CAT tool, working seamlessly on both Mac and PC, that combines multiple machine translation engines, over a dozen specialized terminology tools, and the robust assistance of ChatGPT for Translators. It provides professional translators with advanced functionalities for translation, definition, rewriting, and much more.
Try now! Log in or sign up to receive a free trial subscription!
Visit pro.wordscope.comTrusted by thousands of professional translators in 40+ countries
Visit pro.wordscope.com to create a translation project.