Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture and can integrate different model providers through extensions. If you need to use Command Code models (such as DeepSeek V4 Pro/Flash) and manage them through DSH’s TUI or Web interface, the commandcode-go-for-dsh plugin provides a direct integration solution. It is ported from commandcode-go-for-pi and adds usage query functionality.

Plugin Overview

This is an unofficial plugin and is not part of the official Command Code or DeepSeek projects. It runs as a Provider for DeepSeek Harness and supports invoking Command Code models in the TUI (dsh --profile tui) and Web UI (dsh web).

Core features include:
* Model routing: Forwards model requests to https://api.commandcode.ai/alpha/generate.
* Usage query: View plan balance and reset cycle via the /cc-usage or /ccusage host command.
* Model catalog: Seeds the official Command Code catalog, including models such as DeepSeek, Kimi, GLM, and Qwen.
* Reasoning control: Supports passing reasoning effort parameters in requests.

Installation and Enablement

When installing the plugin, it must be added separately for each Profile in use. The plugin and configuration are isolated, but credentials are shared.

# Install in TUI Profile
dsh plugin --profile tui add github:gonegirl07/commandcode-go-for-dsh

# Install in Web Profile
dsh plugin --profile web add github:gonegirl07/commandcode-go-for-dsh

After installation, the application must be restarted or the browser refreshed for the changes to take effect.

Authentication Configuration

Before using this plugin, configure the Command Code API Key.

Method 1: Environment variable

export COMMANDCODE_API_KEY="user_..."

Method 2: DSH credentials file
Configure it in ~/.dsh/.credentials.yaml; note that file permissions should be set to 600:

COMMANDCODE_API_KEY: user_...

Model Selection and Configuration

The plugin has the official catalog built in, so you can usually select models directly from the interface.

  • TUI operations:

    1. Enter /model to open model selection.
    2. Select Command Code Go.
    3. Enter s to select the first Pro model (for example, deepseek/deepseek-v4-pro).

  • Web operations:

    1. Go to Settings -> Models, or look directly in the Composer model selector.

  • Default model configuration (optional):
    Add to ~/.dsh/settings.yaml:

    agent-default-model:
      provider: commandcode
      model: deepseek/deepseek-v4-pro
      reasoningEffort: high

Usage Example

Enter the following command in the TUI prompt or Web Composer to view usage:

/cc-usage

This command is a DSH host command (Host Slash Command). If it does not appear in the menu, restart dsh web and refresh the browser.

Output example:

Command Code  individual-go
Month   $8.77 / $10 left
5-hour  $0.14 / $3    reset Aug 16, 19:15
Week    $1.23 / $6    reset Aug 22, 10:44
Cycle   ends Sep 15, 2026

Reasoning Level Support

DeepSeek V4 Pro / Flash supports only specific reasoning levels; other levels are not forwarded to the API.

DSH parameter setting API request parameter (params.reasoning_effort)
off The parameter is not sent
high "high"
max "max"

Note: When set to off, Pro models still perform reasoning; they simply do not specify an effort level.

Notes

  1. Unofficial: The plugin is maintained by the community and is not an official project.
  2. Endpoint changes: The /alpha endpoint used by the plugin is not publicly documented and may change at any time.
  3. Network restrictions: dsh web listens only on 127.0.0.1; other devices on the local network cannot access it directly unless using an SSH tunnel or specific port binding.
  4. Closed-source models: Some closed-source model IDs may require a Command Code Pro or Max subscription tier.
  5. Other reasoning levels: Besides off, high, and max, other reasoning level configurations are ignored.

More Information