Redcode v0.69.0

0.69.0

Minor Changes

• Connecting an OpenAI-compatible endpoint now asks for the API base URL and the key, and nothing else. The models and their context and output limits are read from the endpoint's /models (now also from the fields vLLM, LM Studio, OpenRouter and llama.cpp use), and the provider ID and display name come from the host, with a suffix if that ID is taken. The provider ID, display name, API, extra headers, model IDs and limits moved behind one "Customize the connection?" question that defaults to no.

Patch Changes

• The model picker has a Refresh action (ctrl+r) beside Connect an integration and Favorite. It reloads the location's services, which asks every connected provider for its models again, then reads the list back into the picker.
• While the satisfaction indicator has not read enough prompts to show a stage, the footer now shows a quiet mood 1/3 count in dual reasoning, so you can tell it is there and what it waits for. A count that stays at zero means System One is classifying your prompts without enough confidence to read a reaction from them.