gpt-researcher-keenable

Keenable web search as a GPT Researcher retriever plugin. It works without an API key.

pip install gpt-researcher-keenable
export RETRIEVER=keenable

That's the whole setup. GPT Researcher (0.16.1 or later) finds the plugin via the gpt_researcher.retrievers entry point, so nothing in GPT Researcher changes. It combines with built-in retrievers like any other name, e.g. RETRIEVER=keenable,arxiv.

Configuration

Behaviour

• Results come back as {"title", "href", "body"} with requires_scraping = True: body is a single-line preview capped at 500 characters, and GPT Researcher scrapes each page for its content, the same as with the built-in web retrievers.
• max_results from GPT Researcher (MAX_SEARCH_RESULTS_PER_QUERY) is passed to the API, clamped to its 1-50 range.
• With query_domains, the search is restricted to the first domain.
• Any network, HTTP or parsing failure, and an invalid KEENABLE_API_URL, is logged and yields [], so one provider cannot abort a research run.

Development

uv sync
uv run pytest
KEENABLE_LIVE_TESTS=1 uv run pytest -k live   # hits the keyless endpoint

License

MIT