RabbitSoftware Public corpus and local files

Public web research / RabbitSoftware

rabbit-web

Search public sources. Inspect the evidence. Read a bounded AI summary.

JavaScript must load before an approved search can begin.

JavaScript is loading. Nothing is sent while you type.

Skills, tools, and limits

Skills: compare evidence, distinguish findings from hypotheses, explain uncertainty, and cite numbered sources. These are model instructions, not a guarantee of accuracy.

Tools: one Tavily basic search per approved question, up to five linked snippets, followed by one existing-model request. No full-page fetching, autonomous loops, local-file access, code execution, or original application tools.

The site reserves at most 900 search attempts per UTC calendar month and five per network per rolling hour. Failed provider requests still count. Other use of the same Tavily account can exhaust its free allowance sooner. Keep paid overages disabled in Tavily; this site cap does not control account billing or existing model-hosting costs.

The model scales to zero after 15 idle minutes to control hosting costs. Waking can take several minutes and return HTTP 503. There are no automatic retries or keep-alive requests. After a failed summary, approve a manual summary-only retry to reuse these sources without another Tavily search. Changing the question requires a new search.

Pattern checks cannot detect all private information. Do not submit identifying, genomic, biosignal, or personal health data. Search snippets can be incomplete or misleading; open sources to check claims. Not medical advice. Read website privacy.