GooodWei/context-vista

context-vista

See your context window at a glance — token usage, compaction savings, and cost estimates, all in one view.

A /context slash command for DeepSeek Harness that shows the current context token usage and allocation as a donut chart. A mini floating card also stays on the right side of the conversation area, showing occupancy and estimated cost live — draggable and collapsible.

Installation

npx @deepseek-ai/dsh plugin --profile web add github:GooodWei/context-vista
npx @deepseek-ai/dsh web

Install pnpm first (dsh plugin add invokes it internally). If dsh is installed globally, npx @deepseek-ai/dsh can be shortened to dsh.

Usage

Type /context and press Enter to see a card: a context composition donut (system / tools / messages / free), an occupancy bar, session totals, and an estimated cost.

The mini donut card on the right updates live with no input; drag its title bar to move it vertically (position is remembered), and click the arrow to collapse/expand. A "Compact context" button at the bottom of the card triggers /compact (same effect as typing the command); it shows "Compacting…" and disables while the compaction runs.

Custom pricing

Built-in DeepSeek official pricing (CNY ¥, with peak/off-peak). To override or add, edit ~/.dsh/settings.yaml (hot-reloaded, no restart). Currency and peak/off-peak definitions both follow the API (route) — each route can set its own.

Minimal example

context-vista:
  pricing:
    "https://api.deepseek.com":
      models:
        deepseek-v4-pro:
          peak:    { hit: 0.3, miss: 9, output: 27 }
          offpeak: { hit: 0.15, miss: 4.5, output: 13.5 }

Full example (DeepSeek ¥ + OpenAI $, all keys annotated)

context-vista:
  pricing:                                  # pricing table: outer key = route name or baseURL
    "https://api.deepseek.com":             # route 1: DeepSeek, CNY + Beijing-time peak/off-peak
      currency: "¥"                         #   this API's currency symbol (omit for default "¥")
      timezone: 8                           #   timezone of peak/off-peak windows (UTC offset; omit for default 8)
      peakWindows:                          #   peak windows, HH:MM, inclusive start, exclusive end; start > end = crossing midnight
        - start: "09:00"                    #     peak start
          end: "12:00"                      #     peak end
        - start: "14:00"                    #     second peak start
          end: "18:00"                      #     second peak end
      models:                               #   models under this API (required)
        deepseek-v4-pro:                    #     model name
          peak:                             #       form 1 (peak/off-peak split) · peak price
            hit: 0.3                        #         cached input, per 1M tokens
            miss: 9                         #         uncached input, per 1M tokens
            output: 27                      #         output, per 1M tokens
          offpeak:                          #       off-peak price
            hit: 0.15                       #         cached input
            miss: 4.5                       #         uncached input
            output: 13.5                    #         output

    "https://api.openai.com/v1":            # route 2: OpenAI (ChatGPT), USD + no peak/off-peak
      currency: "__P_README_SECTION__quot;                         #   this API's currency symbol
      models:                               #   no timezone/peakWindows → flat pricing
        "gpt-5.5":                          #     model name
          hit: 2.5                          #       cached input, per 1M tokens
          miss: 10                          #       uncached input, per 1M tokens
          output: 30                        #       output, per 1M tokens

Prices are per 1M tokens in each API's billing currency; currency only changes the displayed symbol, with no rate conversion. For APIs without peak/off-peak, just omit timezone / peakWindows. The numbers above are illustrative.

• The outer key can be a provider route name (e.g. deepseek-official) or a baseURL (quote keys containing :).
• Each route can set currency / timezone / peakWindows (each overrides the global default); models is required.
• A model price entry is either flat { hit, miss, output }, or peak/off-peak { peak, offpeak }.
• * is supported as a wildcard for the route name or model name.
• Unmatched models use the built-in default (model names containing flash use the flash tier, otherwise pro).
• The amount is an estimate, not a bill.

License

MIT