v0.2.20

GhostPen v0.2.20

• The link summarizer: paste a URL in the summary window and GhostPen reads the page and streams the summary as Markdown with the active AI profile.

• Three summary depths: Brief (a TL;DR paragraph plus a few key points), Standard, and Detailed (fuller sections, more quotes, the document's numbers kept).

• The model service's /metrics: requests, tokens and speed of the built-in model's prompt and generation, the last request and the one running now. Chat completions honor stream_options.include_usage.

• The built-in runner remembers the split that loaded and reuses it across restarts.

• Built-in models: a successful load no longer crashes the runner while saving its remembered plan (flash attention "auto" is the enum's -1, which @intcast to an unsigned type refuses). Since the plans feature, every GPU load loaded the model and then died as "could not be loaded".

• Settings reads the built-in models' trained context from the GGUF header directly, without loading a model.

• The summary's text column is centered in the window; the menu window is transparent.

• Oriel v0.9.2.