blink
Natural-language codebase search: ensemble filesystem walkers plus Jev relevance scores in a Bun CLI.
Overview
blink (github.com/ellipsis-dev/blink) from ellipsis-dev is a Bun CLI that accepts a natural-language query and directory, optionally walks recursively, and runs an ensemble of filesystem walkers whose hits Jev scores into a ranked table. README example finds auth files in a sample tree with percentage columns per path. Setup requires Bun 1.3.14 plus TYPESAFE_API_KEY. This listing upgrades the thin catalog row with refreshed star count and a full ProductProfile distinct from web search demos.
Problem: Ripgrep and filename search miss intent when engineers ask where authentication or billing logic lives in plain language.
Built for: Developers with a TypeSafe key who want local codebase search via parallel filesystem walkers scored by Jev probabilities.
First indexed on Jev Directory: 2026-09-24
Creator and team
- Name
- ellipsis-dev
- Handle
- @ellipsis-dev
How Jev is used
- Role in the product flow
- Relevance scoring over walker-selected file paths for a natural-language query
- Primitives
- ScoreChoice
- State in
- Query string, target directory path, walker count, and candidate file nodes from parallel tree walks per README CLI flags.
- Decision out
- Ranked file paths with percentage relevance for terminal table output.
- Parse query and directory from CLI arguments
- Spawn multiple filesystem walkers with -n_walkers and optional -r recursive mode
- Collect candidate file nodes from the ensemble
- Jev scores each node against the query
- Print sorted table of paths and percentages
blink keeps search local: walkers explore the tree while Jev supplies calibrated relevance instead of embedding prose summaries. Compare superagents-lab-jev-search and mfm-jev-search for HTTP retrieval over the public web or a YouTube catalog; blink targets your checkout on disk. The ensemble pattern spreads walker diversity before a single scoring pass, which is a useful template when you outgrow ripgrep but do not want a vector database yet.
Sourced performance claims
- README documents Bun 1.3.14 plus requirement and example ranking auth files at seventy four and sixteen percent.Source: github.com/ellipsis-dev/blink README
- Public GitHub repo ellipsis-dev/blink had about seventy three stars when this listing was drafted.Source: GitHub star count September 2026
Features and stack
Features
- CLI with query, directory, recursive, and walker count flags
- Ensemble filesystem walkers
- Terminal table output with percentage column
- Bun-native install path
Stack
- Bun
- TypeScript
- TypeSafe Jev
Pricing: Open source CLI; TypeSafe usage bills to your TYPESAFE_API_KEY per README.
Links
FAQ
- Does blink search the web?
- No. README documents local directory search with filesystem walkers, not HTTP indexes.
- How is blink different from jev-search?
- jev-search plans Search1API web queries. blink scores files inside a path you pass on the command line.
- What runtime does blink need?
- README requires Bun 1.3.14 or newer and export TYPESAFE_API_KEY for live scoring.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.