keenable-haystack

Keenable web search + page fetch for Haystack 2.x, as two components:

• KeenableWebSearch — searches the web and returns documents + links, the same output shape as Haystack's built-in SerperDevWebSearch / SearchApiWebSearch, so it is drop-in for pipelines wired to those.
• KeenableFetcher — fetches a list of URLs and returns documents whose content is the page's main text as markdown (Keenable extracts it server-side, so you don't need a separate LinkContentFetcher + HTMLToDocument step).

Keyless by default: with no API key the keyless public endpoints are used. Provide a key to use the authenticated endpoints (for higher rate limits).

Install

pip install keenable-haystack

Usage

from haystack_integrations.components.websearch.keenable import KeenableWebSearch
from haystack_integrations.components.fetchers.keenable import KeenableFetcher

# No key -> keyless public endpoints. Set KEENABLE_API_KEY to lift limits.
websearch = KeenableWebSearch(top_k=5)
hits = websearch.run(query="latest developments in AI agents")
print(hits["links"])

fetcher = KeenableFetcher()
pages = fetcher.run(urls=hits["links"][:2])
print(pages["documents"][0].content)

In a pipeline (drop-in for any web-search component):

from haystack import Pipeline
from haystack.components.builders import PromptBuilder

pipe = Pipeline()
pipe.add_component("search", KeenableWebSearch(top_k=5))
pipe.add_component("prompt", PromptBuilder(template="Answer using:\n{{ documents }}"))
pipe.connect("search.documents", "prompt.documents")

KeenableWebSearch.run accepts optional per-query filters (site, published_after/before, acquired_after/before, mode). There is no max_results: the API returns a fixed-size result set; top_k (constructor) trims it client-side.

Configuration

• API key (optional). api_key=Secret.from_token(...) / the default Secret.from_env_var("KEENABLE_API_KEY", strict=False). Blank/unset → keyless public endpoints. Serializes by env-var name, never the key value.
• Endpoint (optional). KEENABLE_API_URL overrides the base URL (HTTPS enforced; plain http only for loopback). The endpoint is never a component argument the model can set, so it cannot be used to redirect requests.

KeenableFetcher rejects non-http(s) schemes and private/internal hosts client-side before sending, and (like LinkContentFetcher) skips failed URLs by default — set raise_on_failure=True to surface errors instead.

License

MIT © Keenable