dsh-model-usage
DeepSeek Harness 插件：在会话标题中显示模型使用情况和账户余额——OpenCode Go 计划使用情况（滚动周期/每周/每月）及所选模型的每月上限，以及 DeepSeek 账户余额——仅在匹配的模型处于当前选择状态时显示相应信息。
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:itchenshi/dsh-model-usage#5fb7297931e99a7c2590ea28fe4d4857cabe0852Model Usage 以 dsh-model-usage 发布，当前版本为 0.3.1。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Shows model usage / account balance in the DeepSeek Harness session header, gated on the session's active model selection:
| Active model route | Widget shows | Source |
|---|---|---|
opencode-go / opencode | OpenCode Go plan usage — rolling / weekly / monthly percentages + reset time, plus the selected model's monthly cap (上限 $60) | GET https://opencode.ai/zen/go/v1/usage + per-model limits table (see below) |
deepseek-official | DeepSeek account balance — total / granted / topped-up | GET https://api.deepseek.com/user/balance |
┌─ session header ─────────────────────────────────────────────────────────┐
│ My conversation title [OpenCode Go 滚动 18% 周 82% 月 42% 上限 $60] 打开功能 ▾ │
│ Another conversation [DeepSeek ¥110.00] 打开功能 ▾ │
└──────────────────────────────────────────────────────────────────────────┘
slot: conversation.session.header.actions
The widget registers at conversation.session.header.actions, which the engine
documents as "Title-adjacent Session actions in ascending order" — verified in
the engine's own header source as:
titleRow
├─ titleCluster
│ ├─ crumbs ← the session title
│ └─ headerActions ← THIS widget (right of title)
├─ headerUtilities ← "open feature" buttons
└─ headerCorner ← right-sidebar ExpandButton
The widget renders only when the session's current model routes to a tracked
provider. The selection is read from
ctx.modelDirectories.directoryFor(sessionId).store — the same shared state the
model selector and composer seat use, so a model switch shows/hides the widget
immediately.
Which route belongs to which section is configuration, owned by the host
half: the host returns its live provider→section map (sections) on every
poll, and the client gates on that. Changing providers in cordis.patch.yml
therefore takes effect without touching client code.
| Half | File | Runs in | Job |
|---|---|---|---|
| Host | lib/index.js | Node | Fetches each section from its upstream and serves it over one same-origin JSON route. |
| Client | client/client.js | Browser | Registers the header widget, polls the host route, gates on the active model. |
API keys never reach the browser: the host half resolves them through
ctx.credentials by reference (OPENCODE_GO_API_KEY, DEEPSEEK_API_KEY) and
calls the upstream.
这条路由受引擎的信任围栏保护（Host 白名单 + 浏览器会话 cookie）。它返回账户用量与余额， 因此不带 cookie 的裸
curl会得到401 unauthorized（修复前返回 200 —— 任何本机进程、 以及被 DNS rebinding 的页面都能读到）。同源的页面请求会自动带上 cookie；手工调用需先用dsh web打印的 URL 换一次会话 cookie，再以-H 'cookie: …'传入（示例见dsh-gui-last-session/README.md）。
GET /model-usage
{
"ok": true,
"sections": {
"opencode-go": { "providers": ["opencode-go","opencode"], "keyRef": "OPENCODE_GO_API_KEY" },
"deepseek": { "providers": ["deepseek-official"], "keyRef": "DEEPSEEK_API_KEY" }
},
"opencode-go": { "ok": true, "usage": { "rolling": {…}, "weekly": {…}, "monthly": {…} }, "fetchedAt": 1700000000000 },
"deepseek": { "ok": true, "balance": { "isAvailable": true, "infos": [ { "currency":"CNY", "total":"110.00", "granted":"10.00", "toppedUp":"100.00" } ] }, "fetchedAt": 1700000000000 },
"limits": { "deepseek-v4.1-flash": { "hours5": 12, "weekly": 30, "monthly": 60 }, … },
"limitsMeta": { "source": "docs|cache|builtin", "updatedAt": "2026-09-20T…" }
}
The per-section entries are keyed by section key — the same keys as sections, which is what the page half indexes with (the camelCase opencodeGo / deepseek names are the config keys in cordis.patch.yml, not the wire keys).
Each section reports its own outcome, so a user with only one of the two keys
still gets that half; the other half degrades to a diagnostic label
({"ok":false,"reason":"no-key|unauthorized|network|timeout|bad-payload"})
instead of hiding the widget.
OpenCode Go usage
GET https://opencode.ai/zen/go/v1/usage
Authorization: Bearer <OPENCODE_GO_API_KEY>
200 {"usage":{"rolling":{"status":"ok","percent":18,"resetsAt":"..."},
"weekly":{...},"monthly":{...}}}
DeepSeek account balance (official docs)
GET https://api.deepseek.com/user/balance
Authorization: Bearer <DEEPSEEK_API_KEY>
200 {"is_available":true,
"balance_infos":[{"currency":"CNY","total_balance":"110.00",
"granted_balance":"10.00","topped_up_balance":"100.00"}]}
The host normalizes and clamps the usage payload (percent 0–100) and normalizes the balance amounts (the upstream sends decimal strings; finite numbers are accepted too). Each section is cached for 60s and re-polled after failures no sooner than 30s.
OpenCode Go's usage endpoint is account-wide — it ignores per-model params, and there is no API for the per-model monthly caps documented under "使用限制" (e.g. DeepSeek V4.1 Flash $60/mo, DeepSeek V4 Pro $15/mo). So the plugin:
BUILTIN_MODEL_LIMITS, last synced 2026-09-20),https://opencode.ai/docs/zh-cn/go/ — server-rendered, no JS, no API
key) and parsing the limits + model-id tables into model id → monthly $,$DSH_HOME/logs/model-usage-limits.json,Refresh schedule: at plugin start (background), then every 24h. The widget shows
the selected model's monthly cap (上限 $60) next to the account-level
percent bars, and its tooltip lists the derived caps (5h = 20% · week = 50% ·
month = 100%), the data source (OpenCode Go 文档（9/20） / 本地缓存 /
内置表), and a note that the percentages are account-wide while the cap is
per model. A rate-limited bucket is shown in red with its reset time.
The derived tiers are computed host-side (deriveLimitTiers); the client just
looks up limits[modelId].
dsh plugin --profile web add dsh-model-usage
Restart dsh web (or reopen DSH GUI) afterwards. The same package can be installed
from DSH GUI's Settings → Third-party plugins, or one-click from the
plugin marketplace.
The plugin row lives in cordis.patch.yml; every key is optional:
- insert:
- id: model-usage
name: dsh-model-usage
config:
enabled: true
opencodeGo:
baseUrl: https://opencode.ai/zen/go/v1 # upstream root
apiKeyRef: OPENCODE_GO_API_KEY # credential reference
providers: [opencode-go, opencode] # routes treated as OpenCode Go
deepseek:
baseUrl: https://api.deepseek.com # upstream root
apiKeyRef: DEEPSEEK_API_KEY # credential reference
providers: [deepseek-official] # the engine's DeepSeek route
A top-level baseUrl / apiKeyRef / providers is still read as the
opencodeGo section (this plugin used to be OpenCode-Go-only).
node --check lib/index.js
npm test # local behaviour tests (no network, no engine)
tests/e2e.cjs is a full integration check against a real engine + browser
(it needs a running profile, so it is not part of npm test).
MIT