A general assistant.
Same class of model.
LADLE runs on Claude Sonnet with a 200K context window, web search with cited sources, file uploads up to 30 per chat, images, PDFs, and code in 40+ languages. Streaming word-by-word.
Ask a question. Fund a meal.
- 01You ask.
- 02It streams.
- 03It cites.
- 04It runs code.
- 05A meal lands.
Type a real question, in real words.
No template. No prompt-engineering priesthood. LADLE handles ambiguity the way a good colleague would: asks a clarifying question if needed, then gets on with it.
The first sentence arrives while the last is still writing.
Word-by-word streaming from Claude Sonnet. You read as it thinks. No spinner in front of an answer that’s already ready.
When it searches the web, it shows you where.
Inline citation chips link to a right-hand source rail. Every claim traces to a URL. If we can’t cite it, we don’t say it.
Explains SQL, drafts Python, renders tables.
Syntax-highlighted, copy-ready. 200K context means a whole schema dump fits. Explanations sit alongside, not below.
$0.80 for every ten conversations you have.
At the end of the month, $8 of your $20 becomes ten meals through the World Food Programme. Confirmation reference published the day the donation clears.
200K tokens fits a 10-K.
Most 'AI can read long documents' claims come with a footnote. LADLE doesn't. The full 200,000-token window is available on every chat — roughly 300 pages of English prose, one full annual report, a decently-sized codebase.
Search that shows its work.
When you turn on search mode, LADLE queries the live web and returns answers with inline citations that link to a right-hand rail of sources. If we can't cite it, we don't say it. If a citation exists, that URL was really retrieved and used — verify.
Drop it, ask it.
PDFs, images, spreadsheets, code files — up to 30 attachments per chat. The composer accepts drag-and-drop and preserves file structure (PDFs stay paginated, spreadsheets keep rows-and-columns). No chunking-and-summarizing behind your back.
Reads schemas. Writes migrations. Explains stack traces.
Syntax-highlighted code blocks with a plain-word COPY affordance. Optimized for reading long snippets on a laptop. 40+ languages. For unfamiliar codebases, paste the file plus its two most-related callers — the style match is meaningfully better with context.
Grouped chats with shared context.
A Project holds a set of chats + attached files + an optional voice guide. Every chat in the Project sees the shared context without you re-uploading. The move from 'ephemeral scratchpad' to 'assistant that remembers who you are' is one setup step per topic.
The other half of the product.
Every capability above is one axis. The other axis is the receipt: $8 of every $20 subscription becomes ten meals through the UN World Food Programme, published monthly with a reference number. It's not a marketing sticker — it's an accounting entry with a public URL.
Same class of product. One extra line.
LADLE. One product. One price. One promise.
Cancel anytime. Ten meals a month, minimum, published.