dsh-chat-cost

When juggling sessions across DeepSeek, OpenAI, Anthropic, Google, and other providers in the dsh Web UI, the biggest blind spot is cost. Which subagent burned the budget? Did a session that crossed the peak window get billed at double rate? Without a plugin, the only way to find out is to dig through logs.

dsh-chat-cost puts the answers right in the interface: a hover tooltip on every chat splits spend into the chat itself, its subagents, and the full session tree; each finished answer carries its own per-turn cost badge in the transcript; the composer readout refreshes in sync. Pricing is drawn from a bundled snapshot of seven providers covering 139 models, with DeepSeek peak and off-peak tiers and the published cache-read and cache-write rates for every provider applied exactly as stated. Models without a published price show a dash, never a fabricated number. All spend is logged as append-only JSONL records in the .dsh-cost/ folder inside the project, and interval pricing ensures a chat spanning a peak boundary is billed at the tariff in effect for each segment, not retroactively re-priced.

Six budget tools round out the toolkit: look up per-token and cache rates, pull historical P50/P90 spreads, price a list of work units, pack them under a money cap, mark plan units for clean attribution, and compare routing strategies on quality and cost. No network calls, no telemetry, no accounts. Built for developers running multi-provider LLM workloads in dsh who want to see, at a glance, exactly what each turn costs and how much budget remains.