Introduction¶

DeepSeek Harness (DSH) breaks down model integration, sessions, and tool calls into composable plugins. If you have already subscribed to OpenDesign’s Go plan but still have to manually assemble gateway addresses, credentials, and model lists in DSH, the maintenance cost falls on you every time you switch models or accounts. The community plugin dsh-opendesign-go registers the Design Plan models under the Go plan as unified providers, routes inference through the OpenAI Chat Completions protocol, and centrally handles quota, usage, and multi-account management on the settings page. The following sections explain it in the order of “what it is → what it can do → how to install → how to use”.

What Is This¶

dsh-opendesign-go is a DSH client-side plugin released by maintainer EliteOtaku, with the npm package name matching the repository name. The current released version is 0.1.9 (MIT License). It targets users who have purchased OpenDesign Go subscriptions: it registers a model provider named dsh-opendesign-go on DSH desktop >= 0.2.0-rc.1, exposes the 12 models included in the plan, and comes with usage statistics, quota reading, and multi-account switching capabilities.

Directory page: SkillHub - dsh-opendesign-go
Source code and documentation: GitHub - EliteOtaku/dsh-opendesign-go

Core Features¶

The feature table in the README can be summarized into the following categories, all of which can be configured in the settings page or the profile configuration.

The documentation emphasizes: usage amounts are estimated based on the base tier of the official public price list; long-context tiered pricing may be underestimated, and actual billing is subject to the OpenDesign console. The point of multi-account is “multiple Go subscriptions”; however, multiple Keys within the same workspace still share a single quota.

Model List¶

The Go plan provides 12 models by default (parameters can be overridden in the settings page; sources are listed in the repository’s docs/models.md).

The default inference gateway base URL is https://amr-link.open-design.ai/v1, which can be modified in the profile’s baseURL (no query strings or fragments allowed).

Installation and Activation¶

The plugin requires DSH desktop >= 0.2.0-rc.1 (see engines.dsh in package.json). The Node side requires ^22.19.0 || >=24.0.0.

Desktop (Recommended)¶

In the DSH desktop plugin management interface, enter the npm package name:

dsh-opendesign-go

After installation is complete, restart the desktop app and open the “OpenDesign Go” section in settings.

The README explicitly states: the desktop profile is exclusively managed by the Electron application; do not use the command line dsh plugin add to install it for desktop, otherwise the bundle may fail to resolve or the plugin may not load.

Non-desktop profiles (Command line)¶

For example, the web profile:

dsh plugin --profile web add dsh-opendesign-go

Upgrade to the latest released version:

npm install -g dsh-opendesign-go@latest

You can also enter github:EliteOtaku/dsh-opendesign-go in the plugin management, but the documentation recommends npm: versions and build artifacts are locked, and since npm 12, git dependencies require explicit permission by default. The repository has committed the lib/ build artifacts, so neither path requires local compilation.

Configuration and Credentials¶

Configuration is written under the profile entry opendesign-go, namespace llm-opendesign-go; fields are volatile() references, taking effect immediately after changes without requiring a restart.

The following can be changed directly on the settings page: the enable switch, quota source, account list (including workspaceId), model toggles, per-model parameters, and reconciliation baseline. Others such as apiKeyEnv, baseURL, requestTimeoutMs, autoSwitch, etc. need to be edited in the profile’s config block; complete documentation is available at docs/configuration.md.

Common default values:

API Keys are stored in the DSH credential service: add an account in the “Accounts” card on the settings page (reference name such as OPENDESIGN_ACCOUNT_<n>), or set OPENDESIGN_API_KEY in the startup environment.

When “Use local OpenDesign login credentials to read quota” is checked, the plugin looks for vela in the official installation directory (Windows / macOS / Linux paths are listed in the README), which can also be overridden with the environment variable OPENDESIGN_VELA_PATH. vela is an official private CLI; if its output format changes, the plugin reports an error instead of displaying incorrect numbers.

Typical Usage¶

After the installation and configuration above, the typical workflow is as follows.

1. Prepare the OpenDesign API Key (or multi-account list) in the credential service or environment variables.
2. After restarting or refreshing the profile, select a model from the Go plan in the session model selector (e.g., deepseek-v4.1-flash or kimi-k2.7-code).
3. Enable the quota card as needed: when the OpenDesign desktop app is installed locally, check the vela channel and view the remaining 5-hour / 7-day window in the settings page.
4. When image recognition or tool-based image reading is needed, choose a model that declares the image modality; the plugin places images from tool results into the immediately following user message sent to the model.
5. View tokens and estimated costs aggregated by model on the settings page; when alignment with the console is needed, use “Align current period” to write to alignBaselineJson.

Multi-account scenario: add an accounts entry for each independent Go subscription and configure their respective apiKeyEnv; after enabling autoSwitch, failed single accounts can be retried in list order.

Compatibility and Usage Notes¶

The maintainer documented compatibility tests for host versions 0.2.0-rc.1, 0.2.0-rc.2, and 0.2.1-rc.2 (npm run test:hosts) in the README, along with hands-on test notes on the gateway’s million-token context, 10 image-recognition models, quota, and vela specifications. CI covers Node 22.19 and 24 on Windows / Ubuntu, and includes a standalone hosts job.

Who it’s for: Developers who have purchased the Go plan on OpenDesign and want to centrally select models, view usage, and manage multiple subscriptions within DSH.

Note before installation: DSH plugins run with the current DSH process permissions; before installing, you should read the repository source code and the MIT License to confirm that credential and gateway behavior aligns with your security policy. The community directory SkillHub has no official affiliation with DeepSeek / High-Flyer, and the plugin is also not an official app store entry.

Recent versions (e.g., 0.1.9) fixed issues such as context reservation and max_tokens over-length detection, images being discarded by middleware in the tool chain, and installation timeouts caused by frequent settings writes; for details, see the “Recently Fixed Bugs” table in the README and the CHANGELOG.

Conclusion¶

dsh-opendesign-go brings OpenDesign Go’s 12 Design Plan models, usage, and quota views into the DSH settings page, reducing redundant configuration when building providers yourself. For installation instructions and field semantics, refer to the GitHub README and docs/; plugin summaries and links can be viewed in the community directory.

• Directory page: https://www.skillhub.cn/plugins/EliteOtaku/dsh-opendesign-go
• Repository: https://github.com/EliteOtaku/dsh-opendesign-go