Search & ranking
Score or re-rank candidates from a cheaper retrieval step.
73 rows are filed under this pattern, 73 of them with code. Rows are listed as the catalogue orders them: official first, then rows with code, then by stars.
Filter and search them in the catalogue →
Rows marked “not Jev itself” or “shadow mode” stay listed, but they are not examples of deciding with Jev.
-
Cookbook: Classifying RAG passages
Scores each retrieved passage, then decides in code which reach the answering model — keeping contradictory ones flagged and dropping ones carrying prompt injection.
-
Cookbook: Line-by-line search
Semantic search over a terms-of-service document: one request scores 218 line ids with a Choice, and a Noul checks whether the document answers at all.
-
Cookbook: Re-ranking
Re-ranks 30-passage BM25 shortlists for 40 legal queries with one question per query-candidate pair, reporting large top-1 and top-10 gains.
-
AutoGPT TypeSafe blocks
Seven production blocks — choice, score, yes/no, ask-many, route, pick-best, filter — with a UTF-8 byte budget, verbatim wire capture and eleven test files.
-
OpenViking: retrieval reranking
One Noul per candidate document in a single batched request, with the yes-probability used directly as the relevance score.
-
FastMCP jev_search transform
Two-stage MCP tool search: a wide Choice coarse-ranks the whole catalogue, then a shortlist gets full descriptions plus one Noul each to decide whether it does the job at all.
-
jcode: memory recall without embeddings
Replaces the whole retrieval stack for memory recall — no embeddings, no BM25, no reranker — with one batched Noul per candidate memory.
-
XERJ
Local-first search engine and MCP server in one Rust binary. Its optional `rerank` stage asks Jev one `noul` per top hit (does `document` answer the query in the state?) and re-sorts by the probabilities. The authors report FiQA nDCG@10 rising from 0.24 to 0.36, about $0.64 for 648 searches.written by a model
-
LanceDB TypeSafeReranker
A vector-database reranker that asks one Noul per result and uses the yes-probability as an absolute relevance score, comparable across queries.
-
no-mistakes: Jev review pre-brief, measured and retired
One Score per candidate file to pre-brief code review — measured twice, then removed: more billed input for essentially no wall-clock gain, and offline replay showed the candidate list could not reach where review findings land.
-
jev-chat-jarvis
An Android reply co-pilot that judges intent, timing and risk from on-screen text, while separate models handle OCR and drafting.
-
jevgrep (dzhng)
CLI and agent skill that finds code by asking what it does: `jg "How are telemetry events recorded?" ./repo` has Jev judge relevance across folders, files and declarations and returns files and verbatim excerpts. The author reports the same 8 of 10 SWE-bench tasks solved at ~30% lower cost.written by a model
-
pg-jev
A real PostgreSQL extension exposing the primitives as SQL functions, so a semantic decision can appear in a WHERE clause over any row type.
-
hippo-memory
Biologically-inspired memory for AI agents. Decay, retrieval strengthening, consolidation. Zero runtime deps, SQLite, MCP. Benchmarked retrieval with an opt-in TypeSafe Jev reranker.
-
jev-mcp
A ready-made judgement toolbox for agents: fact verification, content screening, semantic ranking, classification and extraction as separate tools.
-
jev-search
Jev-driven web search: chooses the recency window and the best query rewrite, then reranks results in batches with one noul each.
-
狗头军师 Jev Chat
Chat-reply assistant (desktop overlay, Android floating assistant, iOS keyboard) built on Jev Android. From the visible chat or a screenshot it asks Jev the literal question, the other person's intent, a danger level, whether to reply now and the best action, then has Jev pick among drafted replies.written by a model
-
vector-graph-rag
Graph RAG that stores entities and relations as vectors in Milvus, no graph database. Two-stage Jev reranking asks a noul per candidate relation against the query, then scores the passages of the top relations; by the author's run, 86.61% average Recall@5 on MuSiQue and 2Wiki, 1,000 questions each.written by a model
-
jev4pg
Natural-language SQL and 41 semantic operators for PostgreSQL. Filtering, extraction, ranking, matching and verification go to Jev in bounded batches with validated answers; a choice option needs 0.55 and a noul 0.8 to count. The authors report 98% accuracy against 92% for pg-jev's request path.written by a model
-
neo4jev
Puts Jev inside a knowledge graph traversal: at each node it decides which edge is most worth following.
-
jev-doc-search
Tutorial for finding the page that answers a question in a long PDF with Jev's `Choice`: one option per page for short documents, and for long ones a walk down a PageIndex tree, one small `Choice` per level over section titles and summaries, so neither the 255-option nor the 32k-token limit is hit.written by a model
-
neurolink
The pipe layer of an AI nervous system: one interface connecting provider neurons to an application, across three inference types — generate, stream, and decide. Decide returns typed, calibrated judgments (boolean/choice/score) via TypeSafe Jev or the open-weights Laya, not text.
-
jev-semgrep
grep by meaning, now named sys1grep: for every line Jev answers a noul on whether the line matches a meaning, such as network failure, and a threshold decides. Meanings combine with AND, OR and NOT, a regex can prefilter lines, and the meaning and the text can be in different languages.written by a model
-
jgrep
grep where the pattern is a description: each line becomes one Jev noul, matched at probability 0.5 by default, judged concurrently and printed in input order, so it works on tail -f. On 994 Hacker News titles the author measured 4.6 seconds and $0.012 for one description.written by a model
-
skillranker
Ranks an agent's skills for the next step using live session context, with Claude Code hooks.
-
jev-shell-history
Fish-style zsh history autosuggestions, ranked by Jev rather than by recency.
-
jegrep
Semantic grep for code with no index or embeddings: Jev nouls judge each path, then the passages read, against your description, each a yes probability so thresholds hold across batches; hits come back as line ranges. The author puts a typical search over a few thousand files at about $0.005.written by a model
-
jevgrep
Code search for agents that lack the symbol name: it splits the authorized repository into fragments, asks a Jev noul per fragment whether it answers your question, and returns the exact excerpts with paths and line numbers, via CLI or local MCP. Remote calls stay off until enabled per project.written by a model
-
warrenduffer
Intraday trading bot for Indian stocks placing real orders, with no paper mode: Jev ranks Nifty 50 names in two stages and decides take-profit and exits; code sizes each trade, sets the stop and halts at the daily loss cap, via Zerodha Kite or Kotak Neo. A replay runs the same loop on a past day.written by a model
-
jgrep (npm: jevgrep)
grep for what code does: one Noul per code chunk, diff hunk or CSV row, printed as file:line hits with probabilities. --diff gates a PR in CI on a rule written in English (exit 0 match / 1 clean / 2 error); --tests lists the test files a diff can affect.
-
milvus-model
A library integrating embedding and reranker models from OpenAI, SentenceTransformers etc for semantic search in vector database.upstream description
-
grev
Unix filters that ask questions instead of matching patterns: `grev 'is a vegan meal' menu.txt` keeps the lines Jev says yes to, with related tools that use a choice's confidence or a path score. Pins jev-1.13.0 because thresholds are tuned per model version.written by a model
-
laya-jev-GraphRAG
Agentic GraphRAG engine using swappable System One decision models (local Laya / cloud Jev). Features a complete 4-phase pipeline (Ingestion, Pre-Retrieval, Traversal, Post-Retrieval) and evaluation across Neo4j, Memgraph, Apache AGE, and Kùzu driven by a custom A* traversal algorithm.
-
jev-reranker
Jev-powered relevance filtering and reranking for RAG in Python.upstream description
-
hermes-jev
Typed System One decisions, ranking, verification, and an opt-in Hermes tool gate using TypeSafe Jev.
-
jev-recall
Retrieve by relevance, not resemblance: filter an AI assistant's memories with TypeSafe's Jevupstream description
-
pi-jev-skill-picker
Rank Pi Agent Skills for the current task with TypeSafe Jevupstream description
-
Cheshi
Jev-powered conversation memory: find past sessions and revisit decisions with original sources. A macOS workspace for OpenAI Codex. Manage AI conversations and agents, explore code with CodeGraph, and work with Git, Ghostty terminals, and Apple Notes in one app.upstream description
-
jev-rag-benchmark
Reproducible benchmark for measuring Jev reranking quality, latency, and cost in RAGupstream description
-
transcript-lens
Explore YouTube transcripts by meaning: a Turkish interface, analysis by Jev, subtitle export and a Vercel deploy recipe.
-
Jev RAG
Indexes local documents with SQLite FTS5/BM25, then asks one Jev Noul per candidate passage and sorts by the returned relevance probability before optionally generating a cited answer.
-
jevql
Semantic SQL for Postgres, powered by Jevupstream description
-
oko
Code retrieval over MCP for Codex, Claude Code and OpenCode: hands an agent the relevant source snippets so it reads less irrelevant code.
-
pi-jev
Jev-assisted file retrieval and request caching for faster Pi workflowsupstream description
-
JevGPT: Thinking Fast and Slow in Autonomous Research
Research agent from arXiv 2610.18272. GPT-5.5 plans searches and writes the answer; in BrowseComp-Plus QA, Jev answers a Choice over numbered source lines (where is the answer) and a Noul (does the region address it) per document region and query, in parallel. GPT gets the original lines.
-
jev-rerank-bench
An independent head-to-head against dedicated rerankers across fourteen datasets.
-
jev-search-rerank-eval
Does a TypeSafe Jev rerank beat embedding search? Graded relevance eval (9,831 pairs, 164 zh/en queries) over the Agent Skills Hub catalog, with the judge-circularity bias measured.upstream description
-
jevsearch
A shadcn/ui ⌘K site-search block: a local keyword pass shows hits at once, then the top 20 go to Jev in one request (a Noul per candidate, a Choice for the best page, a Noul for whether any page answers) and are re-ordered or dropped; keyword order stands if TypeSafe is slow or down.
-
every
Ask a yes/no question of every function in a codebase. Ranked answers in seconds, for cents. Grep whose pattern is a question, powered by TypeSafe Jev.upstream description
-
JevPDF
Searches a PDF by meaning. pdf.js extracts each page's lines in the browser and Jev answers one Noul per line ("does this line answer the query?"), at most 16 lines per request with the page text as shared state. Lines at p ≥ 0.55 are highlighted in probability order; only text is sent to Jev.
-
jev-skill-gate
Cut Claude Code's skill manifest by ~75% with TypeSafe Jev. Scores every installed skill for relevance and hides the rest via skillOverrides — 12,750 → 3,185 tokens on a 217-skill install, for $0.0009 a session.upstream description
-
jev.nvim
Neovim: ask the buffer a question, get a quickfix list. Treesitter splits functions, Jev scores each one, probabilities land as virtual textupstream description
-
llama-index-jev
LlamaIndex reranker + router powered by TypeSafe Jev — typed scores/choices, cheaper than LLM-as-judge.
-
jev-retrieval
Semantic code and document search CLI — BM25 recall + TypeSafe Jev calibrated precisionupstream description
-
pijev
PiJev: a terminal coding agent with Jev in the loop — Jev ranks the repository's files before the first call, picks skills and triages failures; your coding model writes the code. Built on Pi.upstream description
-
typesafe-mod
Claude Code mod that routes decisions to TypeSafe's Jev model: ranks installed skills per prompt, and answers the agent's own this-or-that questions when confident.upstream description
-
askgrep
grep for the questions you cannot write as a pattern. Reads every function instead of sampling a few. Powered by Jev, TypeSafe AI's System One model.upstream description
-
jev-assist
Don't burn your expensive main model on grep-and-guess grunt work — let jev rank the whole repo, and save the main model for reading the right files and writing the right code.
-
jev-nlgrep
Search code and text by meaning with natural-language grep, powered by Jev.upstream description
-
jev-reranker
Rerank, filter, and compress JSON search results with TypeSafe AI's Jev.upstream description
-
jev-starter
Typed, policy-driven decision workflows on top of TypeSafe AI Jev: confidence routing, fallbacks, evaluation, and RAG patterns for TypeScript apps.upstream description
-
clay-jev-people-ranker
Agent Skill and Python workflow for Clay lead scoring, B2B prospect qualification, and people-search ranking with TypeSafe JEV.upstream description
-
agent-seek
Agent Seek — precision web recall for agents. You.com discover + TypeSafe Jev ranking. MCP + REST. Live demo: https://agentseek.devupstream description
-
jevgrep
grep by meaning: pipe in any text, ask a yes/no question in plain English, get only the matching lines. Works behind tail -f, about $0.004 per 1,000 lines, powered by TypeSafe's Jev.
-
jev-bfs
Wikipedia link races with direct Jev ranking and a live terminal display.upstream description
-
Jevflix
Movie recommender: FAISS and BM25 search narrow 4,800 films to a shortlist, Jev reads the request's constraints (runtime, no graphic content) for code to filter on, and one Jev choice over at most 250 candidates picks a film. If unsure, an LLM weighs the top five or asks one question.written by a model
-
jfind
Find files by describing them in plain English: find(1) with a semantic --like predicate, answered by TypeSafe.ai's jev modelupstream description
-
typesafe-as-a-judge
Unofficial community MCP plugin for Codex and Claude Code using TypeSafe Jev for bounded routing, ranking, extraction, verification, and escalationupstream description
-
deepseek-harness-jev (luobosibing2)
DeepSeek Harness plugin with twelve opt-in features using Jev as the judgment model: skill selection and glob-result ranking (one `noul` per candidate), completion and goal checks, drift reminders, interjection routing, tool-output and test-log filtering, and approval of single workspace operations.written by a model
-
jev-orderby-bench
Does ORDER BY over a Jev probability put rows in a defensible order? Independent ranking, calibration and invariant measurements of TypeSafe AI's Jev: passes six pre-registered gates on 360 labeled rows, fails four of six on graded product relevance.upstream description
-
jevcrawl
MCP server and CLI that crawls a website for an agent's question. Per page, one Jev request asks a `noul` for every text block (does it help answer the goal?) and two for every link (leads to the answer, or to a section about it); a last `noul` asks whether the kept excerpts fully answer the goal.written by a model
-
sift
Chrome extension that re-ranks Google results with TypeSafe Jev and folds away sales pages and SEO filler.upstream description