dsh-cost-meter

Agentic coding and multi-step AI workflows silently burn through prompts, completions, and context-cache tokens until the monthly bill lands. dsh-cost-meter embeds a high-precision cost chip right into the DeepSeek Harness conversation header, computing incremental prompt, completion, and cache read/write tokens in real time with no polling or UI stutter. One click on the chip opens an interactive breakdown showing your active UTC peak/off-peak window, per-model per-million-token rate tables, and a cumulative session spend summary all without leaving the chat.

The built-in dual-tier tariff engine tracks DeepSeek official peak windows (01:00–04:00 and 06:00–10:00 UTC), auto-detects the 50% off-peak discount, and permanently anchors each historical charge to the rate in force at execution time so retrospective re-rating is impossible. It distinguishes native endpoints from hosted gateways like OpenRouter, lets you pick currency symbol, exchange-rate multiplier, and display timezone, and warns you the moment a discount window is about to open or your session budget threshold is approaching. Cache savings are calculated and highlighted automatically so you can see exactly how much prompt caching is saving you each round.

If you run DeepSeek Harness for agentic coding, mix multiple models in a single session, or work in long-context tasks and want quiet always-on financial visibility without breaking your flow, dsh-cost-meter is the unobtrusive cost sentinel you have been missing. No extra backend, no polling loop just install, converse, and export a Markdown spend summary when you are done.