Introduction¶

In DSH Web, if you have already added custom Providers and models on the Models page, you need to pay attention to two configurations later: which reasoning levels the model supports, and which thinking format the Provider uses.

gao-gao-zai/dsh-custom-reasoning is a DSH plugin maintained by gao-gao-zai, used to configure reasoning levels and thinking formats for custom pi-ai models. It adds a “Reasoning Levels” section to the DSH Web Settings page, centralizing the configuration entry point into one page.

Below is an introduction to its core capabilities, installation methods, typical usage, and limitations to be aware of.

What is this¶

One-sentence positioning: Configure reasoning levels and thinking formats for custom pi-ai models in DeepSeek Harness (DSH).

The main problem it solves is: when you have multiple custom pi-ai Providers and models, you need to view, adjust, and batch-set the available reasoning levels for each model in one place, while selecting the corresponding thinking format for each Provider.

The plugin license is MIT. The DSH Community Directory is an independent site with no official affiliation with DeepSeek or Huanquan.

Core Features¶

The plugin provides the following capabilities:

Configure multiple reasoning levels for each custom model:
off
minimal
low
medium
high
xhigh
max
Configure thinking formats for each Provider:
Auto-detect
openai
deepseek
openrouter
together
zai
qwen
string-thinking
ant-ling
Add a “Reasoning Levels” section to the DSH Web Settings page.
After configuration is saved, the model selector in the chat interface displays the model’s supported reasoning levels as radio buttons.
Auto-save after modification, no need to manually submit.
Support dragging the mouse to quickly check or uncheck multiple reasoning levels.
Support copying a model’s level configuration and clicking or dragging to paste it to other models.
Use Settings revision when saving to avoid silent overwrites and concurrent modifications.

Installation and Enablement¶

Prerequisites:

DSH Web is installed and can run.
At least one custom Provider and model have been added to the DSH Models page.
Custom model configurations are managed by the llm-pi-ai Settings namespace.

Assembly steps in the README are as follows:

git clone https://github.com/gao-gao-zai/dsh-custom-reasoning.git
cd dsh-custom-reasoning
dsh plugin --profile web add .

Among them, the core assembly command is:

dsh plugin --profile web add .

After assembly, restart DSH as prompted and refresh the page.

Typical Usage¶

First, confirm that the current DSH Web is accessible and that at least one custom Provider and model exist in the Models page.

Open DSH Web.
Go to Settings.
Open the “Reasoning Levels” section.
Select the appropriate thinking format for each Provider.
Check the reasoning levels supported by each model.
Wait for the interface to display “Saved”.

Keep at least one reasoning level for each model; the interface will not allow unchecking the last option.

Copying Level Configuration¶

If you want to apply the same reasoning level configuration to multiple models, you can use the copy configuration feature:

Click “Copy Configuration”.
Click any level tag of a model to set that model as the template.
Click the target model, or drag the mouse over multiple target models while holding the button.
Press Esc or click “Cancel Copy” to exit copy mode.

The copy operation only copies the model’s reasoningEfforts and does not modify the Provider’s thinking format.

Saving and Limitations¶

When the plugin saves configurations, please note the following points:

Only manages custom pi-ai Providers, does not manage built-in DSH models.
Reasoning levels and thinking format options are currently provided by a fixed list within the plugin.
Wait 800ms before saving after modification to merge consecutive operations.
Provider thinking format is written to providers.<provider>.compat.
Model reasoning levels are written back to the corresponding Provider’s models array.
When saving model levels, the complete models array of the owning Provider is written back.
The save request carries expectedRevision, so it won’t directly overwrite if the server configuration has changed.
Concurrent modifications are protected by revision checks, but after a conflict, the page needs to be refreshed to reload the latest configuration.
When saving fails, the page displays the API error.
The plugin does not store Provider credentials or model keys locally.
The current package only contains loadable JavaScript files, no standalone build scripts or automated tests.

Applicable Scenarios and Notes¶

This plugin is suitable for scenarios where DSH Web is used, custom pi-ai Providers and models have been added, and centralized management of reasoning levels and thinking formats is desired.

Before installation, it is recommended to check the source code and MIT license. The plugin performs DSH Settings write operations and runs with the current dsh process permissions, so it is not recommended to use it directly in environments with unknown origins or incomplete review.

If multiple people in the team are modifying the same DSH configuration simultaneously, be aware of revision conflicts. When a conflict occurs, refresh the page to reload the latest configuration before continuing to modify.

Conclusion¶

The value of gao-gao-zai/dsh-custom-reasoning is focused: centralizing the management of custom pi-ai model reasoning levels and Provider thinking formats in the DSH Web Settings, and providing batch operations such as drag-and-drop checking and copying configurations.

gao-gao-zai/dsh-custom-reasoning: Configure the reasoning level for the DSH pi-ai model