Parallel Search for DeepSeek Harness

This plugin makes DeepSeek Harness's built-in web_search tool use Parallel Search. It does not add a new tool or provide web_fetch.

It currently supports @deepseek-ai/dsh@0.1.0-rc.6 on Node.js ^22.19.0 || >=24.0.0. DeepSeek Harness is still a developer preview, so later versions may need a plugin update.

Before you start

You will need:

• pnpm 10 or newer;
• a Parallel API key; and
• a model provider already configured in DeepSeek Harness.

You do not need to install dsh globally. The commands below run the supported version through npx.

Install

First, make your Parallel API key available in the terminal where you will run Harness:

export PARALLEL_API_KEY="your-key"

Then install the plugin into the web profile:

npx --yes @deepseek-ai/dsh@0.1.0-rc.6 \
  plugin --profile web add \
  --allow-build=@parallel-web/dsh-web-search-parallel \
  "github:parallel-web/dsh-web-search-parallel#main"

The --allow-build flag lets pnpm build this plugin and nothing else. For a reproducible install, replace main with a full commit SHA.

Stop any running Harness process, then start it again:

npx --yes @deepseek-ai/dsh@0.1.0-rc.6 web

Check that it worked

Inspect the profile:

npx --yes @deepseek-ai/dsh@0.1.0-rc.6 \
  --profile web --dump-config | \
  grep -E 'searchProvider: parallel|web-search-parallel'

You should see searchProvider: parallel and a web-search-parallel entry. Seeing the built-in web-search-deepseek plugin as well is normal—the searchProvider setting decides which provider handles searches.

For direct request evidence, start Harness with information-level Parallel logging:

PARALLEL_LOG=info npx --yes @deepseek-ai/dsh@0.1.0-rc.6 web

After a web_search call, the terminal should show a successful request to https://api.parallel.ai/v1/search. Review logs before sharing them.

Optional settings

The defaults work without extra configuration. If you want to tune the search, edit ~/.dsh/profiles/web/cordis.patch.yml:

- id: web-search-parallel
  config:
    mode: turbo
    maxCharsTotal: 12000
    maxCharsPerResult: 2000

Setting	Default	What it controls
mode	Parallel's advanced default	turbo, basic, or advanced search
maxCharsTotal	25000	Total excerpt characters returned to Harness
maxCharsPerResult	no limit	Excerpt characters kept for each result

Keep PARALLEL_API_KEY in the environment rather than putting it in this file, which is stored as readable text.

If something goes wrong

• dsh is not found: use the npx @deepseek-ai/dsh commands above.
• Port 3080 is already in use: stop the older Harness process before starting another one.
• The provider is still deepseek-official: restart Harness and check whether your cordis.patch.yml contains a later searchProvider override.
• pnpm reports missing DSH peer dependencies: this warning is expected for this Git install. Harness supplies those packages from the profile runtime.
• Harness answers without searching: the configured model decides when to call web_search. Try a prompt that clearly requires current information.
• The provider is unavailable: confirm PARALLEL_API_KEY is set in the same terminal that starts Harness.

Remove

npx --yes @deepseek-ai/dsh@0.1.0-rc.6 \
  plugin --profile web remove @parallel-web/dsh-web-search-parallel

Restart Harness after removing the plugin.

Development and security

See CONTRIBUTING.md for local development, tests, and the packed-artifact review workflow. Report vulnerabilities according to SECURITY.md.

For help, contact support@parallel.ai.

MIT licensed.