webctl

Overview

Problem: Coding agents without built-in web search burn context on noisy SERP snippets, duplicate URLs, and full pages when only a few paragraphs matter.

Built for: Claude Code, Codex, pi, or custom harness authors who want a shell-first search tool that ranks and dedupes hits with TypeSafe Jev before the model reads them.

First indexed on Jev Directory: 2026-09-20

Creator and team

Name

Kyle

Handle

@dorkitude

X profileGitHub

How Jev is used

Role in the product flow

Per-result relevance Score (0 to 10), optional Noul for custom gates, near-duplicate Noul pairs, and per-chunk relevance when scraping or via skimmer

Primitives

ScoreNoul

State in

Search title, URL, and snippet up to 600 characters per hit; optional full excerpt as content; scrape chunks with query and goal; skimmer passes chunk text or list fields with a required goal.

Decision out

Numeric relevance scores with default cut at 6; optional yes or no via --noul; duplicate pairs collapsed to the best-scored copy; scrape and skimmer keep only chunks above threshold while skimmer retains chunks Jev cannot judge.

1. Query configured search providers with cooldown-aware fallback (~/webctl/cooldown.json on 429 or 402)
2. Pass each result batch to Jev with query and --goal for scoring
3. Dedupe pass one on normalized URL or title; pass two on MinHash LSH pairs sent to Jev in one batch
4. Optional --scrape --filter-chunks on top results; or run skimmer skim or filter on local material
5. Print ranked snippets for the agent harness

Features and stack

Features

• Up to three providers per search with reciprocal rank fusion and shared cooldown ladder (15m to 72h strikes)
• Jev scores each hit 0 to 10 with default min 6; --noul and --rubric overrides
• Two-pass dedupe: URL or title collapse, then MinHash LSH plus Jev same-content judgment
• --scrape --filter-chunks for long pages (~2000 char chunks, overlap configurable in skimmer)
• skimmer skim for URLs, files, dirs, and rg -n pipes with path:start-end citations
• skimmer filter for JSON, JSONL, or line lists; keeps items Jev cannot judge
• Homebrew, apt, npm, go install for webctl; Homebrew, apt, Scoop, go install for skimmer

Stack

• Go
• TypeSafe Jev
• Brave, Exa, Parallel, and other providers per docs/providers.md
• skimmer (sibling repo) for non-search skim and filter

Pricing: Open source CLI; TypeSafe Jev usage bills to your key. Search backends may be free tier, keyed, or IP throttled per docs/providers.md.

Links

• Website
• Repository
• Documentation

FAQ

What is skimmer and how does it relate to webctl?

skimmer (github.com/dorkitude/skimmer) is webctl's chunk filter extracted as its own CLI. Use skimmer skim on URLs, files, or rg output and skimmer filter on JSON or line lists. webctl --scrape --filter-chunks uses the same idea on search results.

Does webctl need a Jev key?

README says searching can run without a Jev key using keyless search endpoints, but Jev scoring and filtering need a TypeSafe key from webctl setup. Without a key you get thinner results and no relevance filter.

How is webctl different from jev-search?

jev-search is a TypeScript Cloudflare Worker that plans Search1API queries. webctl is a Go agent CLI that fuses live SERP providers, scores with Jev, dedupes, and optionally scrapes chunks.

What does skimmer do when Jev cannot judge a chunk?

skimmer README states that when Jev cannot judge a chunk (for example a failed batch), skimmer keeps it rather than silently dropping it.

When should I use --scrape --filter-chunks?

webctl README and benchmarks/README.md: snippets are usually enough; scrape pays on long PDFs, threads, and transcripts, not on short dev-doc facts. Summarize mode can shrink scrape payload further when configured.