DeepSeek Harness Plugin: dsh-ollama-cloud

Introduction¶

DeepSeek Harness extensibility relies on plugins. When connecting to Ollama Cloud, manually configuring the Provider is usually required. The dsh-ollama-cloud plugin simplifies this setup process: it exposes Ollama Cloud directly as a reasoning provider for Harness, supports automatic model-list retrieval via the native API, and provides per-model control of reasoning effort.

What It Is¶

This is a model inference plugin maintained by Asheblog. After installation, it automatically registers the ollama-cloud route as a reasoning provider for DeepSeek Harness. It uses Ollama’s OpenAI-compatible interface for conversations and reads data from the /api/tags and /api/show endpoints to update the model catalog.

Core Features¶

• One-click setup: Once installed, the provider is configured automatically; no Adapter needs to be created manually.
• Reasoning effort control: Supports setting thinking effort (Effort) individually for each model, with options Off, Low, High, Max.
• Self-updating model catalog: On plugin startup, it fetches the latest model list from the configured endpoint. The model list can also be viewed without any Key.
• Web capability integration: Ollama’s web search and fetch endpoints are registered as Harness’ ctx.web provider.
• UI integration: The Models panel in Settings displays an API Key input field and a quota progress bar, and the sidebar displays the remaining quota.

Installation and Enablement¶

1. Add the plugin package using the installation command.
2. Restart Harness.
3. Provide authentication credentials.

dsh plugin --profile web add --force https://github.com/Asheblog/dsh-ollama-cloud/releases/latest/download/dsh-ollama-cloud.tgz

After restarting Harness, provide an API Key for Ollama Cloud in Settings → Models. It can also be configured through the OLLAMA_API_KEY environment variable.

Typical Usage¶

1. In the editor’s model menu, select a model under the ollama-cloud provider.
2. In the same menu, select Effort (Off / Low / High / Max).
3. After sending a prompt, the thinking content is streamed into the conversation log just like regular output.

Notes¶

• Key security: Keys are stored via the credential service in write-only mode.
• Conflict handling: Do not install this plugin and dsh-llm-ollama at the same time. The latter registers the same route and can cause a DUPLICATE_ADAPTER error.
• Automatic model retirement: When an endpoint stops listing a model, that model automatically disappears from the catalog.
• Refresh control: Automatic refresh can be disabled by setting autoRefresh: false.

Conclusion¶

dsh-ollama-cloud simplifies the process of connecting DeepSeek Harness to Ollama Cloud and provides fine-grained reasoning effort control. You can view details in the plugin catalog or get the source code on GitHub.