Connect the model that will edit your server files
VarintKeel needs a model to plan its steps. Use a cloud AI provider through a standard API endpoint, a model that runs on your PC, or the built-in model, which is ready to use without your own endpoint. Then press Test, and give the agent a small task first.
Three ways to connect a model
Cloud AI provider
A cloud AI provider is reached through a standard API endpoint that you enter once in the app. It suits hard jobs that span several files, such as comparing a long log with the settings file and the folder tree. With a cloud model, your prompt and the file text the agent reads go to that provider; with a local model nothing leaves your PC. Your API keys stay on the PC, and secrets in tool output are redacted before they reach the model. A provider may charge separately for its API, so read its terms first.
Model on your PC
A model that runs on your PC suits people who do not want paths, logs or settings to leave the machine. With a local model nothing leaves your PC, so your logs and settings stay at home. You pay for that with your own hardware, and a small local model may handle a one-field edit in enshrouded_server.json better than a long log with many lines. Try a small task first, then give it a harder one.
Built-in model
The built-in model is ready to use without your own endpoint, so it is the shortest path from download to a first task. Open the app, choose your server folder and ask for something small, such as reading enshrouded_server.json and listing its fields. Review the card that comes back. You can add a cloud or a local model later and give both the same task, which shows you which one suits your files better.
What the Test button tells you
The Test button checks the endpoint that you entered. It shows the response time and the context window of the model, which is how much text the model can take in one request. Run it after you paste a new endpoint and again when answers feel slow. A short response time helps when a chat reads many files, and a small context window means that a long log should be read in parts, for example with grep for the first error line. If the test fails, check the endpoint address and your key.
Pick a model for the job
Use a bigger model when a task spans several files, for example when a log, the settings file and the folder tree have to be read together before a fix is proposed. A smaller model is enough for changing one field, such as slotCount, in enshrouded_server.json. The tool run_on_model sends a hard step to a bigger model, so you do not have to change your everyday choice for one difficult job. Start small. Compare the results on the same task before you decide.
Quality varies between models, so the diff is your check, not the confidence of the model. Use Ask first to pause before any action that changes something, read the proposed lines and reject what you do not understand. Plan only suits a first look at an unfamiliar folder, because it writes a plan and runs nothing. Test every change by starting the server.
