DeepSeek Harness (DSH) Desktop has a built-in automatic retry mechanism (defaults to 2 retries) when model requests fail. However, in certain scenarios, you may want to adjust this limit or disable automatic retries entirely. The dsh-model-retry-settings plugin solves this problem by allowing you to directly configure the maximum number of automatic model request retries from the DSH Desktop settings interface, without modifying core code.

Plugin Overview

  • Name: dsh-model-retry-settings
  • Core Value: Directly configure the number of automatic model request retries through DSH Desktop settings.
  • Maintainer: y2zyyr
  • Category: Model Inference
  • License: MIT
  • Installation: npm package installation; pure plugin implementation.

Core Features

The plugin mainly provides the following capabilities:

  1. Native Settings Integration: Adds a configuration option to DSH Desktop under Settings → General for adjusting the retry count.
  2. Global Retry Limit: The configured value applies to all model providers that use DSH’s standard retry strategy.
  3. Supported Values: 0, 1, 2, 3, 5, 10.
  4. Default Behavior: The default value is 2, consistent with DSH’s built-in default (behavior remains unchanged after plugin installation if not configured).
  5. Persistence and Real-Time Effect: The configuration persists across application restarts; after modification, the application does not need to be restarted, and the new configuration applies to the next failed request.
  6. Preserves Existing Logic: It does not change DSH’s internal retryable error classification (such as EMPTY_RESPONSE, RATE_LIMIT, etc.), exponential backoff, jitter, or request cancellation mechanisms.

Installation and Activation

Use DSH CLI to install the plugin into a specified web configuration file:

dsh plugin --profile web add @y2zyyr/dsh-model-retry-settings

After installation, the plugin loads automatically. Since the default value is 2, DSH behavior is fully consistent with the built-in default if no changes are made after installation.

Typical Usage

  1. Open DSH Desktop.
  2. Go to Settings → General.
  3. Find the Maximum Model Request Retries configuration item.
  4. Select the desired value.

Value Semantics:
* 0: Only send the initial request; do not perform automatic retries.
* 2: After the initial request fails, automatically retry up to 2 times (up to 3 requests in total).

Notes and Limitations

  • Retry Semantics: The configured value is the number of retries “after the initial request” and does not include the initial request.
  • Error Classification: The plugin only changes the maximum retry count; it does not change how DSH classifies errors. For example, authentication errors and credential errors are never turned into retries.
  • Interface Dependency: The plugin relies on DSH internal/RC interfaces (such as the agent/request-error retry strategy and settings seam). If DSH undergoes a major version update, these interfaces may change, potentially causing the plugin to malfunction.
  • Security: The current DSH version does not expose arbitrary plugin setting namespaces through the general settings API. This plugin uses a restricted localhost route, exposes only the retry-limit value, and restricts access to local/trusted DSH browser contexts.

Summary

dsh-model-retry-settings provides DSH users with the ability to fine-tune model request retry policies without touching core code. For scenarios that require strict control of the retry limit or force fast-fail behavior, this plugin is a practical tool.