Will it hallucinate a citation?

Not in search mode. If LADLE cites [1] linking to a source, that URL was actually retrieved and used. In non-search mode (when you're asking from its training knowledge), it can occasionally fabricate a citation — always verify anything load-bearing. When precision matters, keep search on.

Can I feed it a whole dataset?

Small ones, yes — up to 30 attachments per chat, and 200K tokens is roughly 25MB of plain text. For anything larger you want a script (or ask LADLE to write one). It's not a data warehouse; it's the interactive assistant that helps you write the query.

Does the meals program create a conflict of interest for policy research?

The World Food Programme is a UN body we donate to, not a source we cite. If you're writing about WFP itself, disclose our relationship the same way you'd disclose any other institutional tie. We have no editorial control over what our subscribers write; our involvement ends at the monthly transfer.

USE CASE · RESEARCH

The assistant that shows its sources.

200K context for whole filings, cited web search that links out, and syntax-highlighted code for the analysis after. Built for people whose reputation depends on getting the citation right.

Start with LADLE Watch it work

MODELClaude Sonnet

CONTEXT200K tokens per chat

SEARCHCited · links out

FILESPDFs, spreadsheets, code

OUTPUTStructured, quotable

PRIVACYNever trained on

READ · WHOLE DOCS

Filings, papers, and reports read end-to-end.

Drop a 10-K, a 90-page academic paper, or a policy PDF. LADLE reads the entire thing before answering — no summarize-and-forget. Ask targeted questions and get answers that cite section numbers.

Attached: Uber's Q1 2026 10-Q. Compare 'take rate' as disclosed here vs Q4 2025's. Note any change in how it's calculated.

SEARCH · CITE

Web answers with sources you can verify.

Every claim LADLE makes from web search links to a source in a right-hand rail. No rewriting, no summarization-that-strips-nuance. The citation is inline; the URL is next to it. If the source is wrong, that's on the source, not on LADLE inventing it.

What's the current state of EU regulation on foundation-model training data? I need cited answers for a policy brief.

ANALYZE · STRUCTURE

Structured output for your workflow.

Ask for a table, get a table. Ask for BibTeX entries, get BibTeX. Ask for a CSV of extracted quotes with page numbers, get that. LADLE's output can be shaped for the next tool in your pipeline instead of being a wall of prose you have to reformat.

From the attached 12-paper reading list, produce a CSV: title, author, year, one-sentence contribution, one-sentence weakness. No prose around it — just the CSV.

WHY THIS EXISTS

The professional research subscription price is $20. LADLE is the same, and $8 of it buys ten meals through the UN World Food Programme every month.

Questions.

Live — LADLE queries the web in real time when you enable search mode, and every result links to the source you can visit. Cache TTLs are short (hours, not days) for aggressive freshness. Some highly rate-limited sources may serve slightly older content; that's noted in the citation when applicable.

Try it before you commit.

Open the demo Start with LADLE — $20/mo

← All use cases