dsh-opencode-go-multi-quota

A DeepSeek Harness plugin that shows OpenCode Go quota for every configured account under the composer, and follows the provider selected in the current session.

中文说明 · DSH plugins

中文简介

在 DeepSeek Harness 输入框下方显示所有已配置 OpenCode Go 账号的剩余额度（5 小时 / 每周 / 每月），并跟随当前会话的 provider 自动切换。

• 从 llm-pi-ai.providers 自动识别账号，不在插件里写死数量。
• 按账号数自适应：1 个账号用 320px 紧凑卡，2 个账号用 360px 双列矩阵，3 个以上用「当前账号 + 其他账号」可展开列表，不横向滚动。
• 当前账号列（或当前账号卡）高亮，限流 标红，每格带剩余、已用和 ↻ 重置时间。
• 凭据只在 Host 进程解析，页面只拿到百分比；不会下发 key。
• 完整中文说明见 README.zh.md。

Features

• Multi-account, auto-detected. Reads llm-pi-ai.providers and picks every provider that looks like OpenCode Go. No account count in the plugin code.
• Follows the session provider. Uses the session-scoped modelDirectories store; switching models swaps the pill immediately.
• Adaptive layout. One account uses a 320px compact card; two accounts use a 360px two-column matrix; three or more use a current-account card plus expandable rows for the others, with no horizontal scrolling. The active account is highlighted; rate-limited cells are marked in red.
• All the facts, compactly. Each cell shows remaining %, used %, and reset time (↻). Same-day resets show only HH:mm; other days show MM/DD HH:mm.
• Per-account cache. The host caches each provider separately for 60 seconds, and one failing account does not break the others.
• No keys in the browser. The host resolves credential refs and calls the usage API; the client only receives percentages.

How accounts are discovered

On every request the host reads the loaded @deepseek-ai/dsh-llm-pi-ai entry and its config.providers. A provider is included when it has an apiKeyEnv and any of these is true:

• the provider id starts with opencode;
• its baseURL contains opencode.ai;
• its apiKeyEnv contains OPENCODE.

The apiKeyEnv value is the credential ref to resolve. Providers with only a stored credential record and no apiKeyEnv are not shown.

If the loader config cannot be read, the host falls back to the two conventional refs OPENCODE_GO_API_KEY and OPENCODE_API_KEY. A config with no matching provider shows “No OpenCode accounts found” instead of an endless spinner.

Requirements

• DeepSeek Harness 0.2.1-rc.5 (or a compatible 0.2.x; see engines.dsh).
• At least one OpenCode Go provider in llm-pi-ai.providers with an apiKeyEnv that resolves through DSH credentials or the environment.
• The standard web client with @deepseek-ai/dsh-client-ui-model-selection; the plugin uses it to follow the session provider.

Install

This package is distributed from GitHub only; no npm release is published. Use the GitHub spec below.

Recommended: open Plugins → Install in the DSH web UI and install from the GitHub spec:

github:jackgucarbon/dsh-opencode-go-multi-quota

The Plugin Manager adds the bundle to the profile and reloads it.

CLI installs only add the package dependency; add the bundle to dsh.profile.bundles yourself:

dsh plugin --profile desktop add github:jackgucarbon/dsh-opencode-go-multi-quota

{
  "dsh": {
    "profile": {
      "bundles": ["@deepseek-ai/dsh-base", "@deepseek-ai/dsh-web-app", "dsh-opencode-go-multi-quota"]
    }
  }
}

After installation, restart DSH if the host half does not reload, then refresh the page.

Host API

The host registers a same-origin read-only route:

Request	Response
GET /plugin/opencode-go-multi-quota	Default provider (first discovered, or opencode-go)
GET /plugin/opencode-go-multi-quota?provider=opencode	One named provider
GET /plugin/opencode-go-multi-quota?provider=all	Every discovered account, plus a detected flag
...&refresh=1	Bypass the 60-second cache

The upstream endpoint is GET https://opencode.ai/zen/go/v1/usage with a Bearer token. It is read-only and does not consume plan quota.

Security and privacy

• Credential values stay in the host process. They are never logged, returned by the route, or sent to the page.
• The only network request is the usage call to opencode.ai; the plugin has no telemetry.
• The quota route is same-origin and read-only, but it is not authenticated and refresh=1 bypasses the cache. Do not expose the DSH web server to untrusted networks without adding a proxy or auth layer in front of it.
• The auto-discovery rule is intentionally broad. If you have a non-Go OpenCode provider with an OPENCODE ref, it may appear and fail; narrow the rule or remove that provider from llm-pi-ai.providers.

Development

npm test

The tests use fake API keys and a mocked fetch; no real credentials are needed. test/client.test.mjs runs the client module against a tiny hook runtime and a fake model directory, so it checks provider switching and the popover without a browser.

Compatibility

Built and tested against DeepSeek Harness 0.2.1-rc.5. The plugin uses the client modelDirectories service and the host loader, credentials, and webServer services; breaking changes to those services will need a plugin update.

License

MIT. Not affiliated with OpenCode or DeepSeek; “OpenCode” and “DeepSeek” are used descriptively.