dsh-laa

DeepSeek bills different rates for peak and off-peak hours, so letting an agent run through the peak window quietly inflates your bill. dsh-laa adds a per-session LAA toggle to DSH: once enabled, the instant a peak window begins every running turn is cancelled and subsequent steps are blocked before a single model request is built, so zero tokens are spent. When the off-peak window opens, the interrupted turn receives a continuation prompt to pick up where it left off, and all deferred inputs are replayed in their original order, nothing lost.

Switching is lightweight: type /laa on or tap the sliding switch in the Web UI. Sub-sessions inherit the parent session setting automatically, so the whole session tree shares one toggle. Peak windows follow the official DeepSeek Beijing-time schedule out of the box and are fully configurable, so official pricing changes never require a code update. To extend peak-hour blocking to a self-hosted route, simply add its route id to peakProviders. A small HTTP surface under /dsh-laa (health, state, mode) lets any script read or change the toggle programmatically.

Built for DSH users with long-running tasks, developers who want zero peak-hour spend without changing their workflow, and automation pipelines that need a single knob to gate many sessions at once. The package is a zero-dependency, pure-ESM module that drops straight into a profile with no extra installs or build steps.