dsh-deepseek-usage

Running long, multi-session inference on DeepSeek Harness leaves account balance and token spend nearly invisible. Developers tend to miss runaway sessions until credits are already gone, because checking spend requires breaking out of the conversation. dsh-deepseek-usage closes that gap by packaging two operational queries as model-callable tools, turning cost awareness into part of the dialogue rather than a separate dashboard.

The plugin exposes two core tools. deepseek_balance hits the official DeepSeek balance endpoint and returns a per-currency breakdown (total, granted, topped-up); when a lowBalanceThreshold is configured, it appends an explicit warning the moment the primary-currency balance falls below that ceiling, giving you an early heads-up before the account runs dry. usage_summary aggregates per-session token consumption from the Harness token meter across all live sessions and returns the list sorted by total tokens in descending order, making it easy to spot outlier sessions and gauge how close you are to a model context-window limit. Both can be invoked directly inside a conversation without leaving the chat.

It is well suited for developers and platform administrators who run multi-session, long-running inference workloads on DeepSeek Harness and want API cost monitoring embedded in their workflow—catching budget overruns or runaway sessions proactively rather than reacting after the fact.