dsh-ui-pricing

Tracking LLM spend across dsh sessions, pinpointing which hours are pricier and which model burns the most tokens has often meant external scripts or back-of-the-envelope math. dsh-ui-pricing brings the entire pricing engine inside the plugin: fill in per-model list prices (per million tokens) in a settings card, sketch a default timeline (say 1.0 during business hours, 0.5 overnight), and toggle a daily exception when a specific day deviates. Everything else - recalculation, re-projection, live aggregation - is automatic.

Three layers do the heavy lifting. The durable cost projection prices every provider-reported usage sample under the multiplier in force at that point in time; later chunks for the same step replace rather than double-count earlier ones, and changing the pricing section re-registers the projection so the whole log reprices under the new rules. The CostLine in the composer dock sums the total spend across all live sessions, carries an estimate badge since the figure is computed from usage and multipliers rather than an invoice, and lets you type a corrected total in place - the delta is stored as manualSpend and later usage keeps accruing on top. A 24-hour strip on the right, painted by multiplier band (discount, premium, or neutral), tracks the wall clock with a marker at the current minute. The timeline editor inside the settings card supports click-to-split, drag-to-move boundaries, and direct multiplier entry, making multi-segment policies easy to shape visually.

Built for deployment admins who manage mixed-model, multi-timezone fleets and want real-time cost visibility without standing up a separate billing stack. Models absent from the pricing table contribute zero to the projection; timeline boundaries snap to whole hours; day exceptions copy the default schedule at the moment of creation rather than referencing it live. These constraints can be worked around by editing the raw settings document when minute-level precision or late-edit propagation is required.