inVectus

Find what’s missing.

Explore a collection. Find areas with limited coverage. Inspect the evidence. inVectus maps your documents, flags potential gaps, and keeps a log of what you decide.

Download inVectus Steal My Code (Legally)

Mapped

Real data: 5,000 PubMed ophthalmology papers, mapped by inVectus
Drag to pan · pinch or ⌘-scroll to zoom · click to open

One workspace, from import to decision.

  1. 01

    Bring your own collection

    Import CSV/TSV, Parquet, or folders of text and Markdown. Map columns and mark items as demand, coverage or neutral.

  2. 02

    Search by meaning

    Compare an idea with your collection, inspect the nearest items, and read an experimental calibrated exists / near / open assessment.

  3. 03

    Explore potential gaps

    Browse coverage gaps, stale coverage and bridges between related regions, each with its evidence and a score breakdown.

  4. 04

    Keep your decisions

    Bookmark gaps, record notes and verdicts, and return to saved work after rescanning.

inVectus showing ranked gaps beside an interactive collection map
Native macOS view of a 5,000-paper ophthalmology collection. Gap scores are collection-level candidates for review, not validated discoveries.

Does this idea already exist here?

Type an idea and see its nearest items, with similarity scores and a verdict calibrated on your own collection. Search can compare every stored vector exactly, so nothing is hidden behind an approximate index.

Verdicts hold for this collection only. They never imply global novelty or legal clearance.

QueryLow-dose atropine for myopia in adults

Open · limited coverage in this collection

  • 0.71Atropine 0.01% and axial elongation in children 2021 · coverage
  • 0.64Myopia progression in young adults 2019 · demand
  • 0.58Orthokeratology versus spectacle lenses 2022 · coverage

Illustrative example

Bring Claude or ChatGPT.

Claude Code, Claude Desktop, Codex and Cursor can search, check ideas, list gaps and suggest decisions in your collection, on your own Claude or ChatGPT plan. Choose which model does future work in each project.

Suggestions wait for you to decide. The agent’s client starts invectus mcp on your computer; inVectus opens no network port and never touches your credentials.

$ invectus connect add claude-code
$ invectus connect add codex
Agent suggestion

G-01 neurons · cortex · colliculus

Confirmed gap

105 demand papers and no review in this collection reaches this region. Worth a scoping search before you commit.

Try it: the model picker from the app. Example suggestion.

Your collection stays on your computer.

Offline by default
Import, embeddings, search, regions and gap scans run locally. Model downloads and PubMed/OpenAlex checks need explicit network access.
No open port
Agents talk to inVectus over stdio. inVectus never reads, stores or sends tokens or credential files.
Inspectable
Scores, calibration quality and supporting items stay visible. Every embedding space records its full identity, so vectors never mix.

Build it from source.

Tested locally on Apple Silicon macOS. Windows and Linux builds are targets, not yet verified. Signed installers are a later milestone.

Build instructions Read the progress log 
git clone https://github.com/rbaltodano/inVectus.git
cd inVectus
npm ci && npm run build
cargo build --workspace --release --locked
npm run tauri dev