ASK / YOUR FILES

Ask your
files.

Drop in documents, ask a question, and get an answer that cites the passages it used. Your files are read in this browser and sent to AnyRoute over TLS with your question. This page saves nothing in your browser.

How your files are handled

Read here. Files are opened in this browser and turned into plain text (.txt, .md, .csv, .json, .html, .docx, .pdf). A PDF is read by a reader that this site serves, in a background worker in your browser, and only when you add one; it asks no other site for anything. File names stay here: the router receives documents called doc-1, doc-2 and so on.
Sent over TLS. The text and your question go to AnyRoute. The router reads them in memory to cut them into chunks, embed them and rank them against your question, and asks a chat model to answer from the best few. It writes none of it to a database, cache or log.
Then to models. The chunks and the question go to the embedding model's provider, and the question and the best chunks go to the chat model's provider. On the attested lane the router sends them only to providers whose hardware attestation it has verified and holds fresh. That shows what code is running, not what it does with your text. On the public lane a provider's documented policy applies.
What is kept. Only what any call leaves: a signed receipt with ids, model, provider, lane, token counts, cost, timing and the SHA-256 of the request and of the response, never their text. A hash does not reveal text, but whoever holds an exact guess of a request can confirm it against one, and a receipt can be read by its id.
On this page. Your key and your files stay in this tab's memory, and are gone when you reload or choose Clear. Nothing is written to browser storage.

The endpoint is POST /api/v1/rag. Read what it does and does not claim.

1Your API key

This needs a prepaid Anyroute API key; per-call payment and blind tokens are not accepted here. The key is kept in this tab's memory only: it is not written to browser storage and is gone when you reload or leave. No key yet? Create one in the dashboard.

2Your files

Drop files hereor choose files

Reads .txt, .md, .csv, .json, .html, .docx, .pdf in this browser. Word files: the body text only, not headers, footers, footnotes or comments. PDF files: the text layer only, page by page; scanned pages are pictures and aren't read.

Before you send

Documents0 of 200Within the limit
Text0 B of 2.00 MiBWithin the limit
Chunks0 of 2,000Within the limit
Embeddings calls0 or more of 64Within the limit

These are the router's default limits for one request (200 documents, 2.00 MiB of text, 2,000 chunks, 64 embeddings calls). Over any of them the router refuses the request with 413 before anything is sent to a provider; a router configured differently says so in its refusal. The embeddings calls shown are a floor: an embedding model with a small context needs more.

3Your question

Question

Lane Attested (default)The chat and embedding models run only at providers whose hardware attestation the router has verified and holds fresh. If a step cannot be served on it, the request is refused rather than sent on a weaker lane. Calls made before a refusal are billed and listed.PublicThe router's ordinary lane. A provider's documented policy applies to your text.Router's choiceThe attested lane when the chat model and the embedding model both have an attested endpoint right now, otherwise the public lane. The answer says which lane it used and why.

Chat model

The embedding model is chosen by the router (and named in the answer). The catalog above is read from this router when the page opens.

Passages

Passages to answer from

Passage size (characters)

The files are cut into overlapping passages of this size (100 to 8000; overlap 15%), the passages closest to your question are ranked in memory, and the best 1 to 20 go to the model.

Nothing is sent until you press Ask.