Swearch

Founder

Research condensation for academics: highlight text on a paper in Chrome, get structured AI analysis scoped to your writing project, and append those notes into a linked Google Doc.

Role: Solo builder
Status: CWS MVP · v0.1.0

Problem

Researchers already live in two places: reading (arXiv, PubMed, journals) and writing (usually Google Docs). Most AI research tools optimize for understanding papers inside their own app. Swearch targets the gap between reading and drafting—capture a claim while reading, structure it against an active project, and land it in the Doc you're already writing.

Architecture

System view

┌─────────────────────┐     ┌──────────────────────┐
│  Chrome Extension   │     │   Next.js Web App    │
│  MV3 + React/Vite   │◄───►│   projects · auth    │
│  panel + sidebar    │     └──────────┬───────────┘
└──────────┬──────────┘                │
           │ chrome.identity / fetch   │ Supabase JS / SSR
           ▼                           ▼
┌──────────────────────────────────────────────────┐
│                   Supabase                        │
│  Auth · Postgres + RLS · Edge Functions (Claude) │
└──────────┬───────────────────────────┬───────────┘
           │                           │
           ▼                           ▼
    Google Drive/Docs API         Anthropic Claude
    (list / read / append)        (Edge Functions only)
           +
       OpenAlex (related papers)

Capture → project → Doc

Researcher          Extension (MV3)         Edge Function           Postgres
    │                     │                      │                     │
    │  select ≥20 chars   │                      │                     │
    │  context-menu act   │                      │                     │
    │────────────────────▶│  inject Shadow panel │                     │
    │                     │  restore session     │                     │
    │                     │─────────────────────▶│  Claude analyze     │
    │                     │◀──── structured JSON─│                     │
    │  Add to project     │                      │                     │
    │────────────────────▶│  JWT + highlight     │                     │
    │                     │───────────────────────────────────────────▶│
    │  Export to Doc      │  Docs API append     │                     │
    │────────────────────▶│  (export role Doc)   │                     │
    │◀──── done ──────────│                      │                     │

Stack

Chrome MV3
Next.js
React
Supabase
Anthropic
Google Docs

What we built

Chrome MV3 extension: context menu (summarize, ask, relevance, claims, add, copy), selection Shadow DOM panel, and Chat / Project / Session sidebar
Project-scoped condensation—highlights, chat, and Doc links hang off the active research_project, not a global chat blob
Google Docs loop: link with Context / Export / Both roles; sync caches Doc text for AI; export appends a formatted block via the Docs API
Hard constraints: Anthropic only on Edge Functions, dual Google OAuth clients, withAuthenticatedSession restore on service-worker wake, on-demand injection (no passive scraping), RLS on every user table

Outcome

Ship-ready MVP for Chrome Web Store submission—privacy policy, packaging, and reviewer fixtures included. Docs today are link + cache for AI context + append-on-export, not a full co-writing editor; the north star is draft-linked research notes. Positioning stays narrow on purpose: SciSpace helps you understand papers; Swearch helps you grow the Google Doc you're writing.