lihuu/dsh-plugins--plugins-dsh-tavily-search ↗★ 0

dsh-tavily-search

Tavily-backed WebSearchProvider for DeepSeek Harness: registers a Tavily search provider into ctx.web so the standard web_search tool can use Tavily — zero main-repo changes. 核心用途是增强 DSH 的联网搜索能力。适合对默认搜索结果不满意、希望接入 Tavily 专业搜索服务的用户。需要配置 Tavily API Key。

Package
dsh-tavily-search
Compatibility
Unverified
Cordis peer range
*
Version
0.1.0
License
MIT
Last updated
Aug 16, 2026

Other repositories with this package name

Install

This plugin has no verified bundle, or compatibility checks failed. Read the repository notes first. Read the full README ↗

Configuration

After installing, three things must be in place for web_search to use Tavily.

1. Mount the plugin row

install.sh adds this to $DSH_HOME/cordis.patch.yml:

- insert:
    - id: dsh-tavily-search
      name: '@local/dsh-tavily-search'

2. Point the web seam at Tavily

The base bundle pins searchProvider: deepseek-official. Override it in $DSH_HOME/cordis.patch.yml:

- id: web
  config:
    searchProvider: tavily

3. Configure the API key

The plugin reads the key from its config.apiKey, falling back to the TAVILY_API_KEY environment variable. The key is never stored in this repo — configure it locally, one of two ways:

Option A — plugin config (recommended). Add apiKey to the plugin row in $DSH_HOME/cordis.patch.yml:

- insert:
    - id: dsh-tavily-search
      name: '@local/dsh-tavily-search'
      config:
        apiKey: tvly-xxxxxxxxxxxxxxxx

Option B — environment variable. Set TAVILY_API_KEY in the environment the harness process runs under (your shell profile, a systemd/launchd unit, a process manager, etc.):

export TAVILY_API_KEY="tvly-xxxxxxxxxxxxxxxx"

Get a key at tavily.com. Keys start with tvly-.

Usage

Once configured, the model's web_search tool automatically routes through Tavily — no code changes. Ask the model to search the web and it will return Tavily results (an optional answer plus a list of source URLs).

To verify it works, restart the harness (however you run it) and ask the model to search something.