Does judging a tool description predict what the tool will return?

No. It scores whether the supplied description and context look helpful for the task; it never runs the tool or sees the unseen result, so treat a high score as worth-a-call, not a guarantee.

Will this save context when my agent already read the text?

No. With paths and URLs the full content goes to Jev without entering the main agent context first, but pasting already-read text through classify pays the cost twice.

What breaks on webpages, PDFs, or login-walled articles?

Fetches handle public HTML and text only, strip scripts, follow up to three redirects, and refuse private targets, PDFs, binaries, JavaScript pages, and logins; a fetched page may still be a challenge page.

How should I treat low-confidence, error, or truncated items?

Look closer rather than filtering them out: results keep input order with per-item errors, truncation flags, resolved URLs, and token usage, and truncation or errors are not evidence of irrelevance.

How do I install Jev Sift and give it a key?

Requires Node.js 20+ on PATH; the checked-in dist/server.mjs bundles its dependencies. Provide JEV_API_KEY (or TYPESAFE_API_KEY) in the MCP server's environment, or a private key file at ~/.config/jev-sift/api-key.

What limits apply to a single classify call?

Up to 50 items and 8 concurrent requests, file and page text capped at 60,000 JavaScript characters with truncation flagged, and web downloads capped at 2 MB with a 20-second timeout.

Does Jev Sift persist my source content or results?

No. The plugin does not persist source content or classification results, and webpage fetches receive no Jev credentials, though the classification content is sent to TypeSafe's Jev API.

CommunityJev Agents & MCP

Jev Sift

MCP server that screens files, public webpages, or inline text against a query and returns compact relevance probabilities, so the main agent opens only promising items.

See highlights

Open github.com

Stars: 47
Language: JavaScript
Published by: Community
Author: kbhuw
Last verified: Oct 2026

kbhuw/jev-sift

User repository on GitHub

View on GitHub

Classify first. Read selectively. A portable agent plugin and MCP tool for batch text classification.

Forks: 2
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Let Jev decide what your agent should look at next. Pass a query and a batch of file paths, public webpage URLs, or text. Jev Sift reads the content, sends it directly to Jev, and returns compact relevance probabilities. Your main agent only opens the items worth a closer look. Tool descriptions work too: pass a tool's description, arguments, and relevant context as text to judge whether calling it would help. This evaluates the supplied description; it does not execute the tool or predict its unseen result.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification Jev for Tool calling

Used inJev for Software & DevTools

Languages: JavaScript

Highlights

Call classify with a query plus up to 50 items using exactly one of text, path, or url per item, with unique ids and order-preserving results.
Swap the query for one to eight typed questions: boolean yes-probability, named choice with distribution, or score on an ordered rubric.
Runs over the native TypeSafe Jev endpoint with a single API key and a checked-in bundled server, with file roots and concurrency settings optional.
Web fetches strip scripts, follow up to three redirects, cap text and downloads, flag truncation, and refuse private-network targets.

Watch out

It judges supplied descriptions and content without executing tools; inline text the agent already read cannot recover context cost. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

eutrya

Standalone terminal agent that puts Jev inside its planning loop, proposing text-model steps and scoring candidates against independent rubrics before a deterministic policy executes one.