gpt-researcher-cloro

cloro retrievers for GPT Researcher. The package adds three retrievers:

GPT Researcher (0.16.1 or later) finds the retrievers through the gpt_researcher.retrievers entry point. You do not change GPT Researcher.

Setup

1. Get an API key at dashboard.cloro.dev. The free plan includes 500 credits each month.
2. Install the package in the same Python environment as GPT Researcher:

pip install gpt-researcher-cloro
export CLORO_API_KEY="your-cloro-key"
export RETRIEVER=cloro            # or cloro_aimode, cloro_bing_aimode, or cloro,cloro_aimode

You can combine the retrievers with built-in retrievers, for example RETRIEVER=cloro_aimode,arxiv.

Configuration

Behaviour

cloro:

• Each result is {"title", "href", "body"}. body is the Google snippet, cut to 500 characters. The retriever sets requires_scraping = True, so GPT Researcher scrapes each page.
• One search returns one page of Google results. MAX_SEARCH_RESULTS_PER_QUERY above 10 gives no more results.
• With query_domains, the query gets site: filters joined with OR.
• Results with an empty link or a Google redirect (google.com/goto) link are skipped.

cloro_aimode:

• The first result is the AI Mode answer (Markdown) under raw_content. GPT Researcher uses this text without a scrape. Its href is the Google AI Mode URL for the query (https://www.google.com/search?udm=50&q=...), so the report cites it as Google AI Mode.
• The next results are the sources that the answer cites. They have no raw_content, so GPT Researcher scrapes them. The retriever sets requires_scraping = False for this reason.
• max_results counts the answer and the sources together.
• query_domains is accepted but not applied.
• An AI Mode answer can take 15 seconds or more. The request timeout is 180 seconds.

cloro_bing_aimode:

• The same result shape as cloro_aimode. The first result is the Bing AI Mode answer (Markdown) under raw_content. Its href is the Bing AI Mode URL for the query (https://www.bing.com/copilotsearch?q=...). The next results are the sources that the answer cites.
• Bing puts the prompt in the page URL, so the limit is 3,500 characters after URL encoding. Non-Latin scripts get fewer characters. cloro rejects a longer prompt with a 400 error, and the retriever gives []. GPT Researcher sub-queries are short, so this limit applies only to a very long query.
• The request timeout is 180 seconds.

All retrievers:

• Each search is one cloro request. With the default MAX_ITERATIONS=3, a research_report makes 5 searches per retriever (1 initial search, 3 sub-queries and the original query). That is 25 credits with cloro, and 30 credits with cloro_aimode or cloro_bing_aimode. Other report types can make more searches.
• A missing key, a network error, an HTTP error (for example INSUFFICIENT_CREDITS) or a response that cannot be read is logged and gives []. One failed provider does not stop a research run.

Development

uv sync
uv run pytest
uv run ruff check .
uv build

The tests replace requests.post with a stub and do not call the cloro API.

License

MIT