dsh-ui-peak-rate
DSH web composer 🔥 2× peak-rate badge — shows when the session's model selection matches a DeepSeek peak-rate route during DeepSeek peak hours.
Peak-rate badge in the composer trailing row
A web client plugin for DeepSeek Harness. It paints a 🔥 2× pill in the composer's trailing input row, just left of the model trigger, when both of these hold:
- The session's current model selection matches the DeepSeek peak-rate policy (see Match rule).
- The current time is peak-priced: a weekday (Monday–Friday, Beijing time) whose UTC hour is inside a DeepSeek peak window (by default
01:00–04:00 UTC or 06:00–10:00 UTC, configurable).
Weekends (Saturday and Sunday, Beijing time) are all-day off-peak per the billing rule effective 2026-08-23, so the badge never shows then.
Otherwise the badge is hidden entirely — no layout cost.
- Clock-only, no side effects. The badge reads the local clock and the existing model-directory store — no model-visible output, no session events, no durable state.
- Weekends never badge. Saturday and Sunday (Beijing time) are all-day off-peak per the billing rule, so the pill never shows then.
- Configurable, fail-loud. Providers, peak windows, and multiplier come from
cordis.patch.yml; invalid windows (start >= end, start < 0, end > 24) fail at load, and peakWindows: [] disables the badge entirely.
- Fresh every 60 s. The component re-evaluates on an interval (cleared on unmount) and immediately when config arrives; hidden until the one-shot policy fetch settles.
For example:
Mon 02:30 UTC (Beijing Mon 10:30, a weekday) + model `deepseek-chat` on `deepseek-official`
→ composer shows: 🔥 2×
Sun 12:00 UTC (Beijing Sun 20:00, a weekend) → badge hidden, no layout cost
Install
dsh plugin --profile web add github:haoliangwu/dsh-ui-peak-rate
Built lib/ is committed, so the git install is one line — no prepare script, no allowBuilds permission. Restart dsh --profile web after install (bundle layer stacks compose at boot).
Configure
All three fields default to the DeepSeek peak-rate policy; override any of them in your profile's cordis.patch.yml (~/.dsh/profiles/web/cordis.patch.yml):
- id: dsh-ui-peak-rate
config:
# Provider ids whose sessions may show the badge (default: ['deepseek-official'])
providers:
- deepseek-official
# Peak windows [startHour, endHour) UTC, left-closed right-open (default: [[1,4],[6,10]])
peakWindows:
- [1, 4]
- [6, 10]
# Peak multiplier vs off-peak rate (default: 2)
multiplier: 2
An empty peakWindows: [] disables the badge entirely (always off-peak). Invalid windows (start >= end, start < 0, or end > 24) fail loud at load.
Disable the badge entirely:
- id: dsh-ui-peak-rate
disabled: true
The host half reads Config (schemastery) and exposes it to the browser half through a Connection RPC channel (/peak-rate endpoint config). No transport carries host plugin Config into the browser context automatically; the RPC channel is the explicit transport.
Match rule
The badge shows when the session's current model selection satisfies both of:
state.current.provider ∈ Config.providers (provider match, default ['deepseek-official']), AND
state.current.model contains deepseek (case-insensitive).
The model-id check narrows the provider list so a non-DeepSeek model routed through a listed provider does not trigger the badge. To match a provider whose model ids do not contain deepseek, neither condition holds and the badge stays hidden — extend Config.providers only when the route actually serves DeepSeek peak-priced models.
How it works
- Peak state is a pure client-side clock fact.
isPeakRate(date, windows) returns true iff the date is a weekday (Beijing time, UTC+8) whose UTC hour is inside one of the configured windows (left-closed, right-open). Weekends are all-day off-peak. The weekend boundary follows Beijing time — the billing rule's stated timezone — so Beijing Saturday 00:00 (UTC Friday 16:00) and Beijing Monday 00:00 (UTC Sunday 16:00) are classified correctly, not by their UTC weekday. The component re-evaluates every 60 s through a setInterval cleared on unmount, and re-evaluates immediately when the config arrives.
- The peak-rate policy is fetched once from the host through a Connection RPC channel. The host half registers
ctx.connection.rpc.handle('/peak-rate', ...), reads Config (schemastery: providers, peakWindows, multiplier) from the profile's cordis.patch.yml, and returns it on the config endpoint. The browser half calls ctx.connection.rpc.call('/peak-rate', 'config', {}) once and exposes the result as a reactive source (useSyncExternalStore); the badge stays hidden until the fetch settles.
- The badge and tooltip render from locale dictionaries with config values interpolated. The badge text is
🔥 {multiplier}×; the tooltip lists the formatted windows and multiplier ({windows} / {multiplier} placeholders, e.g. 01:00–04:00, 06:00–10:00).
- The component reads the session's current selection through
ctx.modelDirectories.directoryFor(sessionId).store (the same instance ModelSelect reads), subscribes via useSyncExternalStore, and applies the match rule.
- No model-visible output, no session events, no durable state. The plugin reads only the clock, the existing model-directory store, and the one-shot policy fetch. The peak schedule is a published 2026 constant; a DeepSeek schedule change requires a plugin update to the default
peakWindows.
Build from source
pnpm install
pnpm build # emits lib/index.js, lib/invariant.js, lib/client.js + sourcemaps
lib/ is committed to the repo so git installs work without a build step. After changing source, run pnpm build and commit the updated lib/.
License
MIT