Keenable for DeepSeek Harness

Web search and page reading for DeepSeek Harness through Keenable, with no API key and no account. It works with whatever model you run in Harness, local models included.

The plugin points Harness's built-in web_search and web_fetch tools at Keenable. It adds no new tools: the model asks for a query or a URL exactly as before, and Keenable answers.

Why

The search provider Harness ships with needs a DeepSeek account or a DeepSeek API key. With a local or third-party model and neither of those, web_search fails. Keenable's public endpoints need no key, so search works on a fresh install:

• web_search returns ranked web results with an excerpt of each page and its publication date, when known.
• web_fetch returns a page as markdown. The page is retrieved on Keenable's servers, so a URL the model chose is never requested from the machine running Harness, and private or internal hosts are refused.

An API key is optional. It only lifts the rate limit described below.

Install

Requires DeepSeek Harness 0.1.7-rc.2 or later and Node.js 22.12 or later.

From Harness: open Plugins → Add plugin, enter @keenable/dsh-keenable, then Install → Enable now.

From the terminal: install into the profile you use, then restart Harness:

npx @deepseek-ai/dsh plugin --profile web add @keenable/dsh-keenable
npx @deepseek-ai/dsh web

Profiles are separate: web, tui and headless each need their own install. Replace web with the profile you run.

Then ask your model to search the web. No further setup is needed.

Limits and the optional API key

Without a key, requests go to Keenable's public endpoints, limited to 10 requests per second and 1,000 per hour per IP address. Search and fetch are counted separately. Everyone behind the same network address shares the limit, so a busy office or CI runner can reach it; the tool error then says so.

A key lifts the limit. Create one in the Keenable console, then either:

• set KEENABLE_API_KEY in the environment Harness starts in, or
• enter it as apiKey in this plugin's settings on the Plugins page.

A key in the settings wins over the environment variable.

Settings

Setting	Default	Meaning
apiKey	$KEENABLE_API_KEY	Optional. Switches both tools to the authenticated endpoints.
maxSnippetChars	500	Excerpt length per search result, 180 to 10,000 characters.
fetchLive	true	Fetch pages live from the source. When off, web_fetch returns Keenable's indexed copy and fails for pages it has not indexed.
maxBodyChars	100000	Maximum characters of page text web_fetch returns. Longer pages are cut and marked as truncated.
baseURL	https://api.keenable.ai	API origin. HTTPS only.

Keeping the local fetch provider

The plugin pins both capabilities to Keenable. To keep Harness's local HTTP fetch provider and use Keenable for search only, override the web row in your profile's cordis.patch.yml:

- id: web
  config:
    searchProvider: keenable
    fetchProvider: http

Both fields are needed: a config patch replaces the row's whole config object.

Privacy

Search queries and fetched URLs are sent to Keenable, even when your model runs locally. Requests identify the plugin with an X-Keenable-Title: dsh-keenable header, which the public endpoints require. Results are external content, and Harness marks them as untrusted data for the model.

Troubleshooting

Symptom	What to check
WEB_PROVIDER_CONFIGURED_MISSING	The plugin is not loaded in this profile. Install it into the profile you run, and restart Harness.
"requests share a per-IP limit"	The keyless limit was reached. Wait, or set an API key.
"Fetching from private/internal hosts is not allowed"	Keenable does not fetch localhost or private addresses. Use the local HTTP fetch provider for those (see above).
Search still asks for DeepSeek credentials	Another patch layer pins searchProvider back. Check the composed profile with npx @deepseek-ai/dsh --profile web --dump-config.

Development

npm ci
npm test            # offline: providers, plugin registration, model-facing tools, manifest
npm run test:live   # real Keenable API, no key

The tests load the plugin into a real Cordis context with Harness's own dsh-web seam and web_search/web_fetch tools. The code is plain JavaScript and needs no build step.

License

MIT