jev-graph-search
CLI and agent skill that shortlist notes in Obsidian or Logseq locally, then have Jev rerank them, returning original passages.
Overview
Jev Graph Search helps an AI agent find useful notes and passages in a local Obsidian vault, a folder of Logseq Markdown, or an optional JSON graph snapshot. It builds a shortlist on your machine, then uses Jev to rank the candidates against your question, and returns the original passage with its source reference so the agent can cite the note it used.
The README reports that on one 8,851-node tax-code graph, Jev reranking raised top-five source recall from 63.9 to 81.8 percent. It runs from the terminal via a pinned npm package with no runtime dependencies or as an agent skill, and setup stores a TypeSafe or OpenRouter key in a private per-user config outside the graph. An offline mode with local lexical ranking and an audit command work on the included example without a key. Releases use npm trusted publishing with provenance.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How well does this candidate note or passage answer the question?(inferred) | — |
How it works
- Build a local lexical shortlist from the notes
- Ask Jev to rank candidates against the question
- Return top passages with source references
- Agent cites the original notes
Strengths
- Reported recall gain from 63.9 to 81.8 percent on one graph
- Original passages returned, not summaries
- Offline mode for trying without a key
Caveats
- Recall gain measured on a single graph
- Sends shortlisted passages to the provider
Getting started
`npx --yes --package=jev-graph-search@0.2.2 jev-graph-search setup`, then `... search "Why did we choose this database?" --input ./ObsidianVault`.At a glance
- Type
- Demo
- Status
- beta
- Language
- JavaScript
- Runtime
- Node 20+ CLI and agent skill
- Has code
- yes
Use cases and integrations
How curated lists describe it
Jev-assisted retrieval and evidence-preserving inspection for local Markdown, Obsidian vaults, and Logseq Markdown graphs.
hellogumbo/awesome-jevJev-assisted retrieval over local Markdown, Obsidian vaults and Logseq, keeping the evidence it used inspectable.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
How relevant each shortlisted note or passage is to the question.
Does it work with Obsidian?
Yes, and with Logseq Markdown and JSON snapshots.
Can I try it without a key?
Yes, with --offline on the included example.
Related projects
MCP server over Apple developer docs and WWDC transcripts that uses hybrid search, then Jev relevance scores to rank results.
MCP serverproductionTypeScriptLocal long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
ProjectproductionTypeScriptDeterministic code-graph context for coding agents on Postgres, AGE and pgvector, with optional Jev or local re-selection of symbols.
MCP serverbetaPythonTerminal tool that finds link paths between Wikipedia articles: Jev ranks each page's outgoing links while Python runs the search.
ProjectbetaPythonAgent skill and Python library for Claude Code and Codex that reranks candidate lists with one Jev yes/no relevance question each.
ProjectbetaPythonSingle MCP tool that indexes a repo and has Jev rank a shortlist so a coding agent reads the right files and lines instead of grepping.
MCP serverbetaPython