Choose the model that reads your listings
CJK Workbench works with a cloud AI provider through a standard API endpoint, a model running locally on your PC, or the built-in model. Your choice decides where the text of your listings goes, so pick it before you open a file. Compare the three below.
Three ways to connect a model
A cloud AI provider
A cloud provider suits long, tangled listings when you want a larger model. Keys stay on your PC. The provider receives the prompt text, including the file text the agent reads, so send only listings you are allowed to share, and remember that the provider may charge separately for its API.
A model on your PC
A local model runs on your own PC and suits listings that must not be sent to anyone. Nothing leaves the machine. Quality and speed depend on your hardware and on the model you install, so try it on a short listing first and compare its explanation with the text.
The built-in model
The built-in model needs no API endpoint and no local install. No account is needed to run the app. Whichever option you choose, prompts go only to the model you picked, so you decide where listing text ends up.
Check the connection with Test
Press Test first. The button checks an endpoint before you rely on it, and it shows the response time and the context window, which tells you how much text fits in one request. A long listing may not fit a small window, so test the model you plan to use and run it again after you change the endpoint.
Match the model to the task
A bigger model suits long, tangled listings and the harder step of following one data structure through many functions, while a smaller one is enough to explain a short snippet. Try the small one first. run_on_model sends one hard step to a bigger model and leaves the rest of the session on your usual choice.
Quality varies between models, and any of them can misread a listing. Read every diff. Use Ask first so that each write waits for your OK, and compare the claim with the listing itself before you accept a notes change.
