Documentation
Use a model that runs on your Mac.
Capper speaks the standard OpenAI-compatible model and streaming completion APIs.
Bionic
Install Bionic, download a chat model, then open Settings and enable its Local Model API. Capper expects the standard endpoint athttp://127.0.0.1:1234/v1.
LM Studio
Install LM Studio, download a chat model, then open its Developer page and choose Start server. Capper uses the same standard endpoint at http://127.0.0.1:1234/v1.
Ollama
Install and open Ollama, then load a chat model using Ollama’s own model library and instructions. Capper expectshttp://127.0.0.1:11434/v1.
Custom servers
Choose Custom server in Settings and enter the base URL for a compatible/models and /chat/completions API. A non-loopback URL sends your instruction and captured context to that server—use only a host you trust.
Memory guidance
When the server reports model weight sizes, onboarding recommends the largest non-embedding model at or below roughly half of physical RAM. The estimate is advisory: model overhead and current memory pressure can still change runtime behavior.