James Carl Sitsit

Concordance

A local-first memory vault that answers from what you saved

Shipped

Concordance library: saved items grouped into Coding and AI, each with a summary, tags, and source link
Concordance item detail: summary, tags, key takeaways, and the user note for a saved paper
Concordance chat: an answer about combining BM25 and vector search with three cited sources

Concordance is a local-first AI memory vault. A Chrome extension captures links, pages, videos, selected text, and notes; a local worker summarizes, tags, and categorizes them; search and grounded chat answer only from what you saved, with every citation linked to its source.

At a glance

Role
Solo: design, build, release
Scope
Chrome extension capture, summarize / tag / categorize worker, hybrid search, grounded cited chat
Constraints
Local-first and single-user: no cloud account, no keys required; AI providers strictly opt-in

The problem

People who save a lot can’t find it later. Bookmarks rot, notes scatter across apps, and asking a chatbot about your own reading gets you a confident answer with no sources.

The approach

Everything stays on your machine. A zero-dependency Node API on the built-in SQLite, a Python worker that leases jobs from the same database, and hybrid retrieval that fuses dense and keyword results. Chat answers only from retrieved chunks, checks every citation, and abstains when the evidence is thin.

Stack

  • Node 22 (node:sqlite)
  • SQLite FTS
  • Python
  • Chrome MV3
  • Qdrant (optional)
  • Ollama / OpenAI / Gemini (opt-in)

Status

v1.0.0 · runs locally by design

Have a project like this?

Tell me what you’re building, or book a short call about a similar system.

Let’s talk