dsh-token-usage-observer

Running DeepSeek Harness, Codex, and OpenCode side by side means token usage is scattered across jsonl files and a SQLite database in separate directories. Knowing how much you burned this week, which requests cost the most, or whether your cache hit rate is improving used to mean hand-parsing logs and writing ad-hoc queries.

dsh-token-usage-observer unifies input (both cached and uncached), cache-write, and output token counts from all three sources, computes cache hit rates, and estimates costs with configurable per-million-token prices. It exposes a usage_stats agent tool that supports filtering by source, date range, category, and grouping dimension. In the WebUI, a Token Statistics sidebar dashboard provides stacked bar charts and a per-session detail list, with prices adjustable on the fly. The plugin only reads local logs and never mutates data.

Ideal for developers and teams that mix multiple LLM clients daily and need to track API spend, compare model or preset costs, and monitor cache efficiency. No local build step is required; install and restart the WebUI to start using it.