# DeepSeek Harness

> Connect DeepSeek Harness built-in web search and page reading with a base URL and Search1API key. Add dsh-s1 only when you need extra tools.

Canonical URL: https://s1.dev/docs/integrations/deepseek-harness

[DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) already includes web search and page reading. Configure its built-in tools to use Search1API; no plugin installation is required.

## Connect built-in web search

1. Open **Plugins → Built-in plugin settings → Web search**.
2. Turn on **Use the new Web Search API**.
3. Set **Search / Fetch API base URL** to `https://api.search1api.com/compat/deepseek`.
4. Paste your Search1API key into **API key** and save.

The next `web_search` or `web_fetch` operation uses Search1API without a restart. Search costs 1 credit per request; fetching a page costs 1 credit per page.

See the [guided setup](https://s1.dev/compatible-apis/deepseek) or the [DeepSeek Search API reference](https://s1.dev/docs/compatibility/deepseek) for parameters, credential behavior and limits.

## Optional: add more tools with dsh-s1

Choose `dsh-s1` when you need news, agentic search, sitemap discovery or trending topics beyond the built-in tools. It is a native plugin that runs in-process against the official [@search1api/client](https://www.npmjs.com/package/@search1api/client) SDK.

> **info:** The optional plugin includes an `s1` skill that teaches the model when to search, crawl, or go deeper.

## Install

Add the plugin to a DSH profile:

```bash
dsh plugin --profile web add dsh-s1
```

For a local checkout, use a path spec after building the package:

```bash
cd /path/to/dsh-s1
npm run build
cd ~/.dsh
dsh plugin --profile web add file:/path/to/dsh-s1
```

## Credentials

1. Create a key in the [Search1API dashboard](https://app.s1.dev).
2. Export it as `S1_KEY`:

```bash
export S1_KEY=...
```

The SDK reads the key lazily, so a missing key fails the tool call rather than plugin activation.

## What your agent can do

- **Live web search** across Google, Bing, DuckDuckGo, Reddit, GitHub, YouTube, arXiv, and more
- **Agentic search** (`s1_ask`) with only `query`; the server chooses sources and filters
- **News** aggregation from multiple sources
- **Page crawling** with clean content extraction from a public URL
- **Sitemap discovery** to map a site's URLs or all public links
- **Trending topics** to surface what is hot right now

Every operation is read-only. Search results include source URLs that the agent can follow up on with a crawl.

`s1_ask` preserves `query`, `intent`, result `source` and `relevance`, and optional
`errors`. See [Ask](https://s1.dev/docs/basic/ask).

## Configuration

Each capability can be disabled independently in the plugin config:

| Key | Default | Description |
| --- | --- | --- |
| `search` | `true` | Live web search |
| `ask` | `true` | Agentic search (five credits, 45-second timeout) |
| `news` | `true` | News aggregation |
| `crawl` | `true` | Page crawling |
| `sitemap` | `true` | Sitemap discovery |
| `trending` | `true` | Trending topics |
| `skill` | `true` | Bundled `s1` skill |

## Credits

Search and news requests cost one credit each; crawling a page costs one additional credit per successfully crawled page. Agentic search costs five credits per request. New accounts receive 100 credits without a credit card. See [Credits and limits](https://s1.dev/docs/essentials/credits-and-limits) for current rates.

## Search filters and result fields

Version **0.2.0** supports all 19 [search engines](https://s1.dev/docs/basic/search) and all six
[news sources](https://s1.dev/docs/basic/news), including Bing China, Yandex, Grokipedia, and Reuters.
`time_range` supports `day`, `week`, `month`, and `year`; `page` supports search
pagination. These filters apply to search/news; agentic search accepts only `query`.

Engine-specific result fields live in optional `metadata`: GitHub search exposes
`metadata.kind`, `metadata.stars`, `metadata.language`, and `metadata.num_comments`;
Hacker News search exposes `metadata.story_url`, `metadata.points`, and
`metadata.num_comments`. Optional `published_date` stays at the result top level.
The HTTP response-level `searchParameters` describes the executed request; DSH
returns the mapped result list. These metadata paths apply to search/news; [trending](https://s1.dev/docs/advanced/trending) keeps its separate response shape.

DSH maps each result's API `link` to `url` and retains `metadata` and `published_date`.

## Source

[github.com/superagents-lab/dsh-s1](https://github.com/superagents-lab/dsh-s1)
· [npm](https://www.npmjs.com/package/dsh-s1)
· [awesome-dsh-plugin](https://awesome-dsh-plugin.com/p/superagents-lab/dsh-s1/)