FyleContext
Private beta

Your documents, as context for any AI

Upload your PDFs, docs and web pages once. Claude and any MCP client pull back only the pages that answer the question, or nothing when nothing does.

No spam. One email when your spot opens.

Only the pages that matter
Vector search finds 30 candidate passages, then a relevance model scores each one. Your AI gets the handful that answer the question, not a context window full of near-misses.
Says “nothing relevant”
A calibrated probability gates every result. When your documents don't cover the question, the answer is empty, so the model says so instead of guessing.
MCP and REST
Add FyleContext as a connector in Claude with one-click OAuth, or call the REST API with a key. Search, list documents and read exact pages, with citations.
Reranking by TypeSafe Jev

Every passage gets a yes-or-no judgment

Vector search is good at finding passages that look similar to a question, and bad at telling you whether they answer it. FyleContext asks Jev, a small decision model from TypeSafe, one question per candidate: does this passage help answer the query? The answer is a probability, so results are ranked by relevance and anything below the bar is dropped.

Right passage ranked first
96% → 100%
recall@1, similarity order vs. Jev order
Off-topic questions answered
0%
no-answer queries that still returned a result
Added latency
~840 ms
p95 for one call scoring 30 passages
How the Jev rerank works

How it works

  1. 1
    Add documentsPDF, DOCX, Markdown, HTML, plain text or a URL. Extracted, chunked page by page, embedded.
  2. 2
    Connect your AIPaste one connector URL into Claude, or any MCP client. Read-only, revocable at any time.
  3. 3
    AskYour assistant searches your files and quotes the page it found the answer on.

Be first in

We're onboarding in small batches while we harden things. Grab a spot.