dsh-context-dashboard

When you drive LLM APIs through DSH, three questions come up again and again: how much context window is left in this session, how much has this conversation cost so far, and is the account balance enough for the next burst. dsh-context-dashboard folds all three into a small ring at the bottom of the sidebar. Collapsed by default it shows just a percentage ring and a token readout; one click expands the full panel, another collapses it back, keeping your workflow uninterrupted.

The panel reads the event stream of the currently selected session and automatically detects the channel billing model, branching its display accordingly. In pay-as-you-go mode it lists the API channel, model name, a context-usage bar with percentage, the accumulated session cost across all four token buckets (input, cacheRead, cacheWrite, output), and the account balance. In token-plan mode it adds the session token quota, rolling N-day usage, and weekly and monthly aggregates. DeepSeek official models get automatic peak and off-peak pricing based on Beijing time zones, and the panel annotates each row with the active rate tier. The channel and model lists are dynamically enumerated from the same source as the DSH model selector, so newly added models can be picked up and priced directly in the settings override table without code changes. Balance queries hit the channel official API; keys are pulled from the credentials store or environment variables on demand, used only for that request Authorization header, never persisted, logged, or rendered in the UI.

It is built for DSH users juggling multiple models and channels who want context budgets and cost tracking within arm's reach, and is especially handy when you need to compare spending across metered and flat-rate channels side by side, or simply want a glance to know whether a long conversation is about to hit the context ceiling.