Introduction

The DSH plugin ecosystem emphasizes “everything is a plugin.” The community plugin directory is an independent site with no official affiliation with DeepSeek or FanHuan.

For agent developers, a common requirement is: wanting to integrate a certain model capability into DSH, but not wanting to apply for or configure an API Key first. dsh-deepseek-web-adapter targets this scenario, wrapping the DeepSeek web version into an API-key-free model provider for DSH.

Below is an introduction to the positioning, verified capabilities, installation methods, typical configuration, and current limitations to pay attention to.

What is it

dsh-deepseek-web-adapter is a DSH plugin maintained by huermi.

It does the following:

  • Allows DSH to access the DeepSeek web version via a local gateway;
  • Automatically starts the local gateway upon plugin load;
  • Recycles the gateway upon plugin unloading;
  • Exposes an OpenAI-compatible API;
  • Uses a local MOCK_LLM_KEY, and the gateway does not validate this value.

Its core object is the DeepSeek web account, not the official DeepSeek API Key.

Core Capabilities

Currently verifiable capabilities in the documentation include:

  1. API-key-free DeepSeek web provider.
  2. Auto-start gateway on load, recycle on unload.
  3. OpenAI-compatible API.
  4. Continuous conversation.
  5. Tool calling.
  6. Model calibration.
  7. Sub-agent concurrency; marked as not fully verified.

Note that this is a developer preview version and has not undergone testing in large-scale real environments.

Verified:
- Login;
- Basic Q&A;
- Expert mode switching;
- Single tool call;
- Gateway auto-start/recycle.

Unverified:
- Multi-round tool loops;
- Long conversation migration;
- Sub-agent concurrency;
- Headless mode;
- Image recognition mode;
- Disconnect recovery.

Installation and Enablement

The plugin runs with the current dsh process permissions; you should check the source code and license before installing. This repository uses the MIT License, allowing free forking, modification, continued development, and redistribution.

First, install the plugin:

dsh plugin --profile web add dsh-deepseek-web-adapter

After loading, the gateway starts automatically. The startup time mentioned in the documentation is about 3-8 seconds. The DSH terminal logs will show:

DeepSeek 网页版网关已监听 5688

Configuring DSH Model Provider

The next step is to tell DSH to use this local gateway.

First, edit ~/.dsh/settings.yaml and add the dsweb provider to llm-pi-ai.providers. An example is as follows:

dsweb:
  {
    displayName: DeepSeek 网页版 (免 API),
    apiKeyEnv: MOCK_LLM_KEY,
    api: openai-completions,
    baseURL: http://127.0.0.1:5688/v1/,
    models:
      [
        { id: deepseek-chat, name: DeepSeek 快速 },
        { id: deepseek-reasoner, name: DeepSeek 专家 },
        { id: deepseek-vision, name: DeepSeek 识图 }
      ]
  }

The baseURL here points to the local gateway’s OpenAI-compatible interface.

Then, edit ~/.dsh/.credentials.yaml and add:

MOCK_LLM_KEY: sk-mock-any-value

MOCK_LLM_KEY can be any value because the gateway does not validate it.

After configuration, select DeepSeek Web in the DSH model selector. Options include:

  • DeepSeek 快速;
  • DeepSeek 专家;
  • DeepSeek 识图.

Login

This package only has the host side, no client card UI, so login needs to be done manually.

Open in a browser:

http://127.0.0.1:5688/login

Then login to chat.deepseek.com. If the login page has “Keep me logged in” or “Remember me”, the documentation suggests checking it.

Note that the current limitation description still describes the login state as an in-memory cookie and prompts that re-login is required after closing the window.

Typical Usage

First, verify if the gateway is available. Execute:

curl http://127.0.0.1:5688/v1/models

This command is used to view gateway logs/status.

If the gateway needs a restart, you can use the uninstall and reinstall method:

dsh plugin --profile web remove dsh-deepseek-web-adapter && dsh plugin --profile web add dsh-deepseek-web-adapter

If you want to start the gateway locally for debugging, you can execute:

node resources/dsweb-gateway.js --port 5688 --base resources/runtime

If you want to run parser regression tests, you can execute:

node tests/test-parser-all.js

Suitable Scenarios and Notes

Suitable for:
- Wanting to integrate DeepSeek web without configuring a DeepSeek API Key;
- Wanting to use the local gateway via an OpenAI-compatible API;
- Wanting to directly select DeepSeek Fast, Expert, or Image Recognition in the DSH model selector;
- Willing to manually complete the browser login and accept the current Beta status.

Needs special attention:
1. This plugin is a developer preview version and has not undergone large-scale real-world testing.
2. The plugin automatically starts the local gateway upon loading and recycles the gateway upon unloading.
3. Depends on Chrome installed on the local machine.
4. The login state is an in-memory cookie; re-login is required after closing the window.
5. Depends on the DeepSeek web UI; UI updates may cause the selector to fail, requiring an update to driver.js.
6. The tool calling parser has a fallback but is not universal; the 54 scenario regression tests cover common formats, but extremely bizarre formats may still fail.
7. This package only has the host side, no client card UI; login must be manually opened at http://127.0.0.1:5688/login, and configuration can be done via curl calling /config. The current documentation does not provide specific request parameters or examples for /config.
8. Requires Node 18+ and Chrome, with no third-party runtime dependencies.
9. The plugin runs with the current dsh process permissions; check the source code and license before installing.

Links

Directory page:

https://www.skillhub.cn/plugins/huermi/dsh-deepseek-web-adapter

GitHub:

https://github.com/huermi/dsh-deepseek-web-adapter