Connect a model for your incident documents
Buncha Build works with a cloud AI provider through a standard API endpoint, a model running locally on your PC, or the built-in model. The choice decides where the prompt text goes, so decide it before you open documents. Version 2.9.30 offers all 3.
Three ways to run the model
A cloud AI provider
A cloud AI provider suits people who already have an account with one and want a larger model. Buncha Build connects through a standard API endpoint. Prompts leave the PC. The prompt text, including any file text the agent reads, goes to that provider, which may also charge separately for its service, so keep sensitive material out of the working folder.
A model running locally on your PC
A model running locally on your PC suits documents that must stay on the machine, such as contact lists and internal notes. Nothing leaves the PC. Prompts go only to the endpoint you picked, which here runs on the machine itself. The trade-off is hardware: a smaller model may write less well, so read every diff and test the result.
The built-in model
The built-in model comes with Buncha Build, so no cloud account or separate endpoint is needed to start. Prompts go only to the model endpoint in use, and only a model running locally on your PC keeps every prompt on the machine. Choose a local model when your documents must not leave the PC. Switch later if needed.
Check the connection with the Test button
The Test button checks an endpoint before you start work. It shows the response time and the context window. The context window matters for documents, because a small one holds fewer pages at once, and a slow response time makes a long review tiresome. Test again after changes.
Picking a model for the task
Use a bigger model for heavy rewrites, such as restructuring a long procedure into phases or reconciling several documents, and a smaller one for small edits like fixing a contact line or a heading. run_on_model sends one hard step to a bigger model while the rest of the chat stays on the smaller one. Try both on a copy.
Quality varies. Read every diff, keep Ask first on while you learn what a model does well, and correct what it gets wrong. No model replaces your own knowledge of your systems.
