Introduction

The DeepSeek Harness (DSH) ecosystem adopts the philosophy that “everything is a plugin.” dsh-connect-trae is a standalone bundle plugin that connects locally logged-in Trae models to DSH, providing usage/credit overviews and daily check-in claim functionality. This plugin is maintained by dingminhua and aims to solve the challenges of unified management and status querying for local Trae accounts within DSH.

Installation and Activation

It is recommended to install using the DSH plugin command:

dsh plugin --profile desktop add dsh-connect-trae

Or install directly via npm:

npm install dsh-connect-trae

After installing, updating, or uninstalling a bundle, you need to restart the corresponding DSH process.

Parallel Dual-Provider Integration

The China and International versions are two parallel providers (trae / trae-global) and can be used simultaneously. The plugin settings card distinguishes them with tabs for easier unified management. Each provider maintains a complete runtime stack—including a credential store, model catalog, selection state, and context budget—without interfering with the other. You can select the China or International version in different sessions, or enable both at the same time.

Auto Detection and Multi-Account Switching

The plugin supports automatic recognition of both the China and International regions and automatically discovers locally installed, logged-in accounts: Trae CN, TRAE SOLO CN, Trae, or TRAE SOLO. The region is automatically determined by the userRegion declaration included with the credentials, so no manual specification is required. The plugin supports multi-account switching, and tokens are not written into DSH settings.

Model Coverage and Tool Loop

Models come from the Trae SOLO channel and include DeepSeek-V4-Flash-Official, DeepSeek-V4-Pro-Official, GLM-5.3, GLM-5.2, Kimi-K3, MiniMax-M3, Qwen3.8-Max, Doubao-Seed-*, and others. DSH obtains structured tool calls through Trae’s llm_utils_chat, executes them with DSH local tools, and then returns the results to the model.

Models from the Trae Code channel (glm-5.3-flash, kimi-k2.8-preview, qwen3.8-flash) are not currently supported. Support can only be added after they are officially made available on the SOLO channel.

Image Recognition and Usage Management

Image recognition capability follows Trae’s own annotation (the multimodal field included in the Trae catalog or manual selection). Only models with multimodal === true automatically enable image input; models with a missing field or one explicitly set to false are treated as text-only.

Except for the daily check-in claim, all plugin queries are read-only and do not consume quota. The check-in claim also only occurs when you click the button. The China version displays Work credits and general credits, while the International version displays subscription/trial status.

System Requirements and Security

Requires DSH 0.1.7-rc.1 or later. The peer upper bound is <0.3.0-0. The plugin uses a secure loopback shim, with one random port per region plus an in-process random secret; the real Trae token is not directly exposed to pi-ai.

Ecosystem Background

DSH’s philosophy is “everything is a plugin.” This plugin is hosted in community directories (such as dshfind.com) and has no official affiliation with DeepSeek or High-Flyer.