Second Brain

An automated knowledge capture system for macOS. It passively logs everything you browse during the day, then generates daily highlight pages in your Obsidian vault. Your browsing history becomes your external memory.

How It Works

Chrome/Comet browser
    |
    v
[Browser Extension] -- passively captures URLs, titles, domains, timestamps
    |
    v  (native messaging)
[Native Host]        -- saves captures directly to local SQLite database
    |
    v
[Generate]          -- creates Obsidian-compatible daily notes
    |
    v
[Fetch]             -- extracts full page content (arxiv API, article extraction)
    |
    v
[Curate]            -- AI generates topic clusters, summaries, [[wikilinks]]
    |
    v
~/Documents/Obsidian/second-brain/2026-04-14.md + morning email digest

Three components:

1. Browser Extension (extension/) -- A WXT-based Chrome/Comet extension that silently tracks page visits with dwell-time filtering (5s minimum), smart blocklist (Gmail, social media, banking filtered out), and a popup UI for pause/quick-block controls.

2. CLI Pipeline (pipeline/) -- A Node.js CLI (second-brain) with seven commands:

   • second-brain export -- Imports a legacy export.json snapshot into SQLite; current extensions sync directly
   • second-brain generate -- Writes Obsidian-compatible markdown daily notes from captures
   • second-brain fetch -- Extracts full page content (arxiv API, article extraction, enhanced meta)
   • second-brain curate -- AI-generates summaries with topic clustering and [[wikilinks]]
   • second-brain capture-conversation -- Logs Claude Code conversation topics to daily note
   • second-brain email -- Sends morning email digest via Gmail/gws
   • second-brain search -- Full-text search across captures and daily notes

3. Shared Types (shared/) -- Zod schemas and TypeScript types shared between extension and pipeline.

Prerequisites

• macOS (launchd scheduling is macOS-only)
• Node.js >= 20 and npm >= 10
• Chrome and/or Comet browser

Quick Start

git clone https://github.com/LEARNableLabs/second-brain.git
cd second-brain
./setup.sh

This installs dependencies, builds the extension, and creates ~/.second-brain/. It then prompts you to load the extension in Chrome:

1. Open chrome://extensions and enable Developer Mode
2. Click Load unpacked → select extension/.output/chrome-mv3
3. Copy the extension ID, then finish setup:

./setup.sh <your-extension-id>

Restart Chrome/Comet, then open the extension popup and click Sync now. Confirm that it says Synced just now. Verify the CLI with npx second-brain --help.

When updating an existing installation, rerun ./setup.sh <your-extension-id> and reload the extension at chrome://extensions. The new background retry alarm requires the alarms permission. If you move the checkout or change your Node installation, rerun setup to refresh the native host launcher.

Configuration (optional)

Configuration lives at ~/.second-brain/config.json:

mkdir -p ~/.second-brain

Output directory — daily notes are written to ~/Documents/Obsidian/second-brain/ by default:

{"outputDir": "/absolute/path/to/your/obsidian/vault"}

AI provider — defaults to claude-code (Claude Code CLI, no API key needed). Alternatives:

{"llm": {"provider": "claude"}}

Requires ANTHROPIC_API_KEY in your environment.

{"llm": {"provider": "ollama", "model": "llama3.1"}}

Fully local and private — requires Ollama running.

Email digest — requires an installed, authenticated gws CLI. By default the digest goes to the signed-in Gmail user; configure another recipient with:

{"email": {"to": "you@example.com"}}

Preview without contacting Gmail:

npx second-brain email --date 2026-09-22 --dry

--to you@example.com overrides the configured recipient. Invalid configuration stops the command, preserving the selected output location and AI provider.

Usage

Browse normally

The extension captures pages automatically. It filters out noise (Gmail, Google Search, social media, banking) and requires a 5s dwell time before logging a visit. Use the popup to:

• Pause/resume capture
• Quick-block the current domain
• See capture stats and local sync status
• Edit skipped domains directly in the popup
• Retry a failed sync with Sync now

Run the full pipeline manually

npx second-brain export                    # Optional: import a legacy export.json snapshot
npx second-brain generate                  # Create daily note
npx second-brain fetch                     # Extract full page content
npx second-brain curate                    # AI summary + topic clusters
npx second-brain curate --eod             # End-of-day polished summary
npx second-brain email                     # Send yesterday's digest via Gmail

generate, fetch, curate, and email support --date YYYY-MM-DD and --dry. Dates use your local timezone. Export supports --dry or --dry-run; previews retain the snapshot for a later import. curate --dry still calls the configured AI provider to generate its preview.

Search your captures

npx second-brain search "transformer attention"
npx second-brain search "react hooks" --from 2026-01-01 --domain dev.to
npx second-brain search "RLHF" --notes-only        # search daily notes only
npx second-brain search "arxiv" --db-only           # search captures DB only

Capture Claude Code conversations

npx second-brain capture-conversation --topic "RL training loop" --summary "Discussed PPO vs GRPO..."

Set up hourly automation

./pipeline/launchd/install.sh

This installs a launchd agent that runs the full pipeline every hour. End-of-day summary at 11 PM, morning email digest at 8 AM. Logs at ~/.second-brain/logs/second-brain.log.

Each daily note includes:

• AI-curated highlights with topic clusters and [[wikilinks]]
• YAML frontmatter (date, capture counts, top domains, browsers)
• Browsing log grouped by domain with meta descriptions
• Source markers (star for manual saves, history suffix for backfills)

The output file is atomically written (safe while Obsidian is open) and auto-committed to a git repo in the output directory.

Project Structure

second-brain/
  extension/          # Browser extension (WXT + TypeScript)
    components/       #   Storage, types, dwell tracking, blocklist
    entrypoints/      #   Background service worker, popup UI
    tests/            #   Capture, sync, and popup regression tests
  pipeline/           # CLI pipeline (Node.js + TypeScript)
    src/
      ai/             #   LLM provider abstraction, curation prompt, vault scanner
      commands/       #   export, generate, fetch, curate, email, capture-conversation, search
      config/         #   Config reader with Zod validation
      db/             #   SQLite connection, migrations, operations, content table
      extractors/     #   Domain-aware content extraction (arxiv, article, general)
      generators/     #   Markdown engine, frontmatter, meta fetcher, atomic writer
      git/            #   Auto-commit to output repo
      messaging/      #   Chrome Native Messaging protocol
      search/         #   Full-text search across notes and captures
    launchd/          #   macOS launchd agent for hourly automation
    tests/            #   CLI, database, messaging, and note regression tests
  shared/             # Shared Zod schemas and types
    src/
    tests/

Development

# Run all tests
npm test

# Run tests for a specific workspace
npm test --workspace=extension
npm test --workspace=pipeline

# Type check
npm run type-check --workspaces --if-present

# Build the extension and check pipeline compilation
npm run build

Verification and troubleshooting

• Local sync unavailable: rerun setup with the correct extension ID, restart the browser, and click Sync now. Captures remain queued in the browser until the native host confirms they are saved.
• No highlights: check the selected AI provider. The default needs an authenticated Claude Code CLI; Ollama needs a running local server and the configured model. A failed curation preserves the previous highlights.
• No email: preview the digest, then check gws authentication and the recipient. Email tests mock Gmail; the test suite never sends mail.
• Custom data location: SECOND_BRAIN_DATA_DIR selects a separate database/config directory for CLI commands and tests. The browser host uses ~/.second-brain by default.

Automated checks cover framed native-host input in a separate process, database persistence, retries and retention, hourly note regeneration, conversation summaries, popup interactions, and email argument construction. Loading the unpacked extension, visual rendering in Chrome/Comet and Obsidian, real AI generation, and Gmail delivery still require a configured local installation.

Roadmap

Built in phases -- see .planning/ROADMAP.md for full details.

• Phase 1: Browser Extension Foundation
• Phase 2: Data Export Pipeline
• Phase 3: Daily Note Generation
• Phase 4: Content Processing (full page content fetching)
• Phase 5: AI Curation (summaries, topic clustering, wikilinks)
• Phase 6: Automation (hourly launchd scheduling)
• Phase 7: Conversation Capture & Email Delivery