dsh-token-usage

When building and running complex AI Agents, Token consumption often remains a black box. It is notoriously difficult to accurately track the hidden costs of context compaction and retries, let alone establish effective budget alerts or compliance audits. dsh-token-usage is designed to solve these exact pain points, offering a local-first, privacy-centric Token observability and trajectory auditing solution for DeepSeek Harness, ensuring every unit of AI compute is fully transparent.

The plugin's core strength lies in its meticulous passive ledger design and multi-dimensional analysis engine. It precisely breaks down four Token buckets—uncached input, output, cache read, and cache write—while independently accounting for context compaction and retry overhead. With features like a 30-week heatmap, robust run-rate forecasting, and anomaly detection, users can effortlessly monitor usage trends and manage rolling budgets. Furthermore, its on-demand AI analysis capabilities stand out: whether generating global usage optimization recommendations or conducting deep trajectory reconciliation for a single session, it strictly adheres to privacy boundaries by relying solely on aggregated metrics or whitelisted metadata, never exposing conversation payloads.

If you are a DeepSeek Harness developer or team manager focused on AI operational costs, requiring fine-grained Agent efficiency tuning, or prioritizing data privacy and compliance auditing, this plugin is an indispensable tool. It operates without intercepting requests or altering routes, empowering you to find the optimal balance between performance, cost, and security through pure, evidence-driven insights.