ycm50/dsh-subscribe-meter#dsh-opendesign-go-meter

DeepSeek Harness session cost meter plugin · OpenDesign Cloud edition (bilingual UI)

A custom fork of dsh-cost-meter that keeps every upstream feature (session cost, official balance, budget box, custom-provider balance, history, peak/off-peak pricing, official price sync, coding-plan quotas, token heatmap, …) and swaps the Go quota card for OpenDesign Cloud subscription quota.

English | 中文

---

Upstream and credits

This plugin is a custom fork of dsh-cost-meter, a complete DeepSeek Harness session cost meter: session cost, official balance, budget box, custom-provider balance, history, peak/off-peak pricing, official price sync, coding-plan quotas, token heatmap, a 90+ model price catalog with automatic matching, and a bilingual UI, maintained by Han-1413141 under MIT.

Why fork: we route models through OpenDesign Cloud (desktop app / Vela CLI / https://amr-link.open-design.ai/v1). Upstream's Go quota card only talks to opencode.ai/zen/go/v1/usage and cannot read OpenDesign quotas. This fork replaces that card with the official OpenDesign quota endpoint and ships OpenDesign's 39-model price catalog.

Credits: the vast majority of this code and design comes from upstream dsh-cost-meter and its contributors — Han-1413141/dsh-cost-meter (MIT). Our changes are the OpenDesign-specific layer described below.

---

What this fork changes

Change	Detail
Quota endpoint	OpenDesign Cloud official quota API in two steps: GET https://amr-api.open-design.ai/api/v1/workspaces/current for the workspaceId, then GET /api/v1/workspaces/{ws}/billing/funding-check for codingPlan.windows[]
Two windows	OpenDesign has only rolling 5 hours (18000s) and 7 days (604800s); both render as "used / total + progress bar". OpenDesign has no monthly rolling window, so the old monthly slot is gone entirely
Units	The API returns Design Plan credits, 1 USD = 10000 credits; values are shown as returned, never reverse-derived
Plan tier	The card also shows the current tier and wallet balance, both taken from the same funding-check response
Credentials	Uses the Vela CLI sign-in controlKey (run npx @powerformer/vela-cli login once; it writes ~/.amr/config.json), not an sk-* inference key. Resolution order: DSH credential store VELA_CONTROL_KEY → env VELA_CONTROL_KEY → ~/.amr/config.json
Price table	Ships an opendesign provider with 39 models, from the public GET /api/v1/public/model-prices endpoint (fetched 2026-10-30, cacheRead included)
Copy	All "Command Code GOAT" text is now "OpenDesign Cloud"; adds a goQuotaTierNote string (zh/en)

Install

Requires Node.js ≥ 20 and DeepSeek Harness with the dsh plugin command.

# Local directory
dsh plugin --profile web add link:./dsh-opendesign-go-meter

# Git URL
dsh plugin --profile web add github:ycm50/dsh-opendesign-go-meter

# dshmarket (if published there)
dsh plugin --profile web add dsh-opendesign-go-meter

Restart dsh web afterwards:

dsh web

lib/ is the shipped artifact — there is no build step (package.json has no build; files lists only lib, cordis.patch.yml and docs/provider-pricing.json). Use npm test (= scripts/verify-typert.mjs) during development.

Compatibility: the strict codecs in lib/typert.host.js and lib/client.js must carry both schema: <zod v4> and create: () => <zod> — the 0.1.5-rc.x host only reads schema while 0.1.7+ only reads create. Missing either key makes the whole typert contribution fail to register on that host generation (all remote methods gone, settings page and cost panel dead). Run npm test after editing lib/.

Sign in: built-in login entry

Quota queries use the Vela sign-in, not an API key. The plugin ships a built-in login entry — no terminal needed:

1. Open Settings → Cost → OpenDesign subscription quota and click "Sign in to OpenDesign";
2. The card shows a 6-character code and an "Open approval page" button;
3. Confirm the code in the browser — the plugin detects completion and refreshes the quota immediately.

Under the hood it launches the official CLI's login subcommand (device authorization); the CLI itself writes credentials to ~/.amr/config.json. CLI resolution order:

1. vela on PATH (if you installed the CLI yourself);
2. the npx cache copy of @powerformer/vela-cli (%LOCALAPPDATA%/npm-cache/_npx/*, run directly, no network);
3. only if neither exists, fall back to npx --yes @powerformer/vela-cli login (needs network the first time).

Signing in from a terminal works too and is equivalent:

npx @powerformer/vela-cli login

The plugin then reads profiles.prod.controlKey from ~/.amr/config.json automatically. The same value can live in the DSH credential store or the VELA_CONTROL_KEY environment variable (handy for CI or multiple machines).

⚠️ There is deliberately no manual key field in Settings: a controlKey outranks an API key (it can read the whole billing ledger and workspace), so it should not be pasted into a text box. To configure it from Settings, store VELA_CONTROL_KEY in the DSH credential store instead.

Quota semantics (measured)

Go is $10/month; the docs say "5 hours $10 / 7 days $30", but those are model-usage dollars. The API returns Design Plan credits (1 USD = 10000 credits), so the real numbers are:

Window	API limitCredits	Shown in the panel	Model usage at 0.5×
Rolling 5 hours (18000s)	50000	$10	$10
7 days (604800s)	150000	$30	$30

The panel shows the official documentation figures $10 / $30, converted at 5000 credits = $1 (50000/10 = 150000/30). This is a different unit from the wallet balance, which is 10000 credits = $1 (see creditsPerUsd in billing/summary). The 2× gap is exactly the reciprocal of the Design Plan multiplier.

Plans (official /model-credits page):

Plan	Price	Design Plan monthly	Plan-model credits	5-hour	7-day
Go	$10	$120	$0	$10	$30
Plus	$20	$160	$20	—	$40
Pro	$100	$360	$120	—	$90
Max	$200	$600	$300	—	$150

Only Go has both a 5-hour and a 7-day window; Plus / Pro / Max only have the 7-day window, and the card automatically shows a single row on those accounts.

These values live in OPENDESIGN_PLANS in lib/coding-plans.js, sourced from the official pages (re-checked 2026-10):

• https://open-design.ai/model-credits/
• https://open-design.ai/help/billing/

Model pricing (the opendesign route)

If models are routed to OpenDesign's Provider API (provider: open-design, baseURL https://amr-link.open-design.ai/v1), the price table must know that provider, otherwise calls record tokens but cost stays 0 ("today's cost" shows ¥0). This fork ships an opendesign provider table with 39 models from the public /api/v1/public/model-prices endpoint, each entry carrying sourceUrl / checkedAt.

• All entries are catalog flat prices per 1M tokens (cacheRead included).
• Model ids use the Provider API spelling (deepseek-v4.1-flash, glm-5.3-flash, kimi-k2.7-code); case and separator differences are absorbed by normalized matching.
• This is the pay-as-you-go catalog price and is a different scale from Design Plan credits: the former converts tokens into USD-equivalents, the latter is the subscription allowance itself.

Edit or mount entries under Settings → Cost → Extended price table. Sync official prices only rewrites the DeepSeek main table; provider tables are checked manually.

Relationship to upstream

• Branch od-go is this fork (OpenDesign edition).
• Other dsh-cost-meter forks (e.g. cmdc-goat) live in their own repositories/branches.
• This branch contains no Command Code / OpenCode Go endpoints or copy.

License

MIT, same as upstream.