dsh-provider-auto
Run the following command in DeepSeek Harness:
dsh plugin install icekale/dsh-provider-auto
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install icekale/dsh-provider-auto in DeepSeek Harness to install the plugin (source: https://github.com/icekale/dsh-provider-auto), then restart dsh web once to activate it.
About this plugin
When you wire an OpenAI-compatible gateway into DeepSeek Harness, the /models endpoint typically returns bare identifiers with no capability metadata. The dsh-llm-pi-ai reasoning selector depends on a declared reasoningEfforts array, so reasoning-capable models like GPT-5, o1, o3, o4, and codex simply have no selector in the UI, leaving users unable to pick or switch effort levels.
This plugin watches the llm-pi-ai settings section and fills in the missing declaration for recognized reasoning model families (gpt-5, o1, o3, o4, other o, codex) with the four standard levels: off, low, medium, high. It skips image, audio, embedding, moderation, and reranking models, and it never overwrites an existing reasoningEfforts declaration. Because it writes to the persisted configuration, DSH native model picker and request path remain authoritative.
Built for developers who route inference through custom OpenAI-compatible gateways and want the reasoning selector to appear without hand-editing capability maps or modifying the gateway itself.
Use Cases
- Reasoning-effort selector is missing when connecting GPT-5, o1, or o3 models via a custom gateway
- OpenAI-compatible /models endpoints return no reasoningEfforts metadata, so the UI selector never appears
- Managing reasoning models across multiple vendors and wanting automatic capability declaration without manual edits
Best For
- Developers routing inference through custom OpenAI-compatible gateways to multiple model providers
- Teams that want the DSH reasoning selector to work out of the box without hand-writing capability maps
- Platform engineers unifying heterogeneous model services inside a single DeepSeek Harness instance
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.