Context Capture Saver

Capture useful context from websites, PDFs, and YouTube transcripts into local JSON, Markdown, and/or SQLite storage you control.

Context Capture Saver is designed for research and AI workflows where you want structured, provenance-aware data instead of ad hoc copy/paste.

Screenshots

Settings Page

Bubble Menu

Key Features

• Capture article-focused page content, not raw HTML.
• Save with notes, highlights, and queued highlight workflows.
• YouTube transcript capture with transcript-only, timestamped-segments-only, or both.
• Bubble menu support on YouTube selections, including transcript save actions.
• PDF extraction with metadata and selection-aware capture support.
• Local storage backends: JSON (default), Markdown, SQLite, and JSON + SQLite (write both).
• JSON output location is configurable (root or json/ subfolder).
• Optional structured JSON chunks (content.chunks) in addition to full documentText.
• SQLite backend uses a graph-ready schema for analytics and AI agent workflows.
• Automatic migration of legacy SQLite databases (captures-only schema) to the latest schema.
• Pending highlights/notes panel now shows queued annotations on-page with delete controls before save.

Browser Support

• Chrome/Chromium (Manifest V3)
• Firefox build output included in the same pipeline

Firefox Differences

• chrome.offscreen is not always available; the extension now detects this and uses module-based fallbacks for PDF extraction.
• Clipboard-assisted fallback paths can be more limited on Firefox; on-page selection remains the primary workflow.
• Runtime capability info is captured in diagnostics for PDF captures when diagnostics are enabled.
• Pending highlight rendering uses CSS Highlights when available, and a DOM-wrapper fallback when it is not.

Firefox Verification Checklist

• Manual:
  • Verify context menus toggle between regular and YouTube actions based on tab URL.
  • Verify PDF capture works with module fallback on Firefox.
  • Verify highlight/note queueing works on regular pages and PDF pages.
  • Verify pending highlight rendering works on browsers with and without CSS Highlights support.
  • Verify JSON, SQLite, JSON+SQLite, and Markdown save modes all write successfully.
  • Verify bubble menu actions work and update after settings changes.
• Automated:
  • Run npm run check
  • Run npm run typecheck
  • Run npm test

Install

Option 1: Chrome Web Store

• Install from the Chrome Web Store

Option 2: Release Artifacts

• Chrome package: context-capture-saver-chrome.zip (and compatibility alias context-capture-saver.zip)
• Firefox package: context-capture-saver-firefox.zip

Option 3: Build From Source

npm ci
npm run build

Build outputs:

• Chrome unpacked extension: dist/chrome/
• Firefox unpacked add-on: dist/firefox/
• Chrome zip: dist/context-capture-saver-chrome.zip
• Firefox zip: dist/context-capture-saver-firefox.zip

Load in browsers:

1. Chrome: open chrome://extensions, enable Developer mode, click Load unpacked, select dist/chrome/.
2. Firefox (temporary): open about:debugging#/runtime/this-firefox, click Load Temporary Add-on, select dist/firefox/manifest.json.

Build Commands

npm run build         # build both Chrome + Firefox
npm run build:chrome  # build only Chrome
npm run build:firefox # build only Firefox

First-Time Setup

1. Open extension options.
2. Choose a local folder with read/write permission.
3. Run Test write (optional but recommended).
4. Configure:
   • storage backend (JSON, Markdown, SQLite, JSON + SQLite)
   • JSON folder organization options
   • compression and diagnostics options
   • logging level (off, error, info, debug) for runtime observability
   • optional JSON structured chunk export
   • bubble menu layout/style/actions

Capture Workflows

Websites

Use right-click menus, shortcuts, popup, or bubble actions:

• Save content
• Save content with note
• Save content with highlight
• Highlight
• Highlight with note

YouTube

On supported YouTube video pages:

• Save transcript
• Save transcript with note
• Selection bubble includes YouTube transcript actions

Transcript storage modes:

• plain transcript text
• timestamped transcript segments
• both

PDFs

• Extracts text + metadata
• Supports capture and highlight actions
• Browser PDF viewer behavior can still affect some selection UX

Keyboard Shortcuts

Default shortcuts:

• macOS: Shift + Command + D, Shift + Command + C
• Windows/Linux: Ctrl + Shift + D, Ctrl + Shift + C

In Chrome, edit shortcuts in chrome://extensions/shortcuts.

Data Storage Overview

JSON Mode

• Default mode.
• Writes one capture file per save.
• JSON files can be written to the selected folder root or a json/ subfolder.
• Optional setting can include content.chunks for agent-ready chunked retrieval while keeping full text.

Markdown Mode

• Writes one .md file per capture with YAML frontmatter metadata.
• Body contains the core document text (no structured chunks or timestamped segments).
• Highlights and notes use footnote-style anchors ([^hl-N]) placed inline where the highlighted text appears, with a "Highlights & Notes" footer section.
• Markdown files follow the same folder organization options as JSON (date/type grouping).
• Markdown is a lossy format compared to JSON/SQLite -- optimized for human readability, not round-trip data interchange.

SQLite Mode

• Writes/updates a single context-captures.sqlite file in the selected folder.
• Uses SQL.js (no external DB server).
• Automatically upgrades legacy DBs in place.
• Runs guarded maintenance (ANALYZE) during schema/open checks at a bounded interval, and stores maintenance metadata in meta.

JSON + SQLite Mode

• Writes both outputs in the same save operation.
• If one backend fails and the other succeeds, successful output is preserved and surfaced.

Privacy

• Data is written only to your selected local folder.
• No required remote backend.
• No cloud dependency in normal capture flow.

Development

npm ci
npm run lint
npm run typecheck
npm run test:ci
npm run build

npm run check dynamically syntax-checks all non-vendor runtime modules under src/.

Contributing

1. Keep changes scoped.
2. Run checks/tests before opening a PR.
3. Include a concise behavior-change summary.

i18n Contribution Flow

1. Add/update message keys in _locales/en/messages.json.
2. Add corresponding translations in _locales/nb/messages.json (or other locale folder).
3. Use translation lookups in runtime code via src/i18n.js (t(key, fallback, substitutions)).
4. Keep sensible fallback strings in code so missing keys do not break UX.
5. Run npm run check, npm run typecheck, and npm test.

SQLite Schema (Graph-Ready)

The SQLite backend is normalized for both analytical queries and AI-agent retrieval.

Core principles:

• One document per source URL.
• One capture per save event.
• Multiple chunks per capture/document (document chunks, highlights, notes, transcript segments).
• Graph tables (entities, edges, provenance) enable relation modeling with evidence links back to chunks.
• Optional derived tables (annotations, transcript_segments) make common querying cheaper without losing the chunk-centric model.

Tables

• meta: DB metadata (db_schema_version, migration/backfill flags, FTS availability, maintenance timestamps/mode/result).
• documents: canonical source objects (url, title/site/language/published metadata).
• captures: immutable-ish capture events (hashes, transcript payload, diagnostics, optional legacy payload).
• chunks: agent-ready text units with offsets/type/provenance anchor.
• annotations: structured highlights/notes derived from capture annotations.
• transcript_segments: structured transcript lines with stable segment ordering.
• entities: graph nodes (document, capture, site, annotation, future extracted entities).
• entity_aliases: alternate labels/mentions for entities.
• edges: graph relations (predicate, confidence, optional properties).
• provenance: evidence links from entity/edge subjects back to chunks.
• chunks_fts (optional virtual table): full-text index when FTS5 is available.
• captures_legacy (optional): preserved legacy table after auto-migration.

Mermaid Diagram

erDiagram
  META {
    text key PK
    text value
  }

  DOCUMENTS {
    text document_id PK
    text url UK
    text canonical_url
    text title
    text site
    text language
    text published_at
    text metadata_json
    text created_at
  }

  CAPTURES {
    text capture_id PK
    text document_id FK
    text capture_type
    text saved_at
    text content_hash
    text document_text
    int document_text_word_count
    int document_text_character_count
    text document_text_compressed_json
    text transcript_text
    text transcript_segments_json
    text diagnostics_json
    text legacy_record_json
  }

  CHUNKS {
    text chunk_id PK
    text document_id FK
    text capture_id FK
    text chunk_type
    text text
    text comment
    text created_at
    int start_offset
    int end_offset
    text source_hash
    int is_preview
  }

  ANNOTATIONS {
    text annotation_id PK
    text document_id FK
    text capture_id FK
    text annotation_type
    text selected_text
    text comment
    text created_at
    int start_offset
    int end_offset
  }

  TRANSCRIPT_SEGMENTS {
    text segment_id PK
    text document_id FK
    text capture_id FK
    int segment_index
    text timestamp_text
    text text
    text created_at
  }

  ENTITIES {
    text entity_id PK
    text entity_type
    text name
    text canonical_id
    text properties_json
    text created_at
  }

  ENTITY_ALIASES {
    text entity_id FK
    text alias
  }

  EDGES {
    text edge_id PK
    text from_entity_id FK
    text to_entity_id FK
    text predicate
    real confidence
    text properties_json
    text created_at
  }

  PROVENANCE {
    text provenance_id PK
    text subject_type
    text subject_id
    text chunk_id FK
    text evidence_text
    text created_at
  }

  CHUNKS_FTS {
    text chunk_id
    text text
    text comment
  }

  DOCUMENTS ||--o{ CAPTURES : has
  DOCUMENTS ||--o{ CHUNKS : contains
  CAPTURES ||--o{ CHUNKS : produces
  DOCUMENTS ||--o{ ANNOTATIONS : has
  CAPTURES ||--o{ ANNOTATIONS : includes
  DOCUMENTS ||--o{ TRANSCRIPT_SEGMENTS : has
  CAPTURES ||--o{ TRANSCRIPT_SEGMENTS : includes
  ENTITIES ||--o{ ENTITY_ALIASES : has
  ENTITIES ||--o{ EDGES : connects
  CHUNKS ||--o{ PROVENANCE : evidences
  CHUNKS ||--o{ CHUNKS_FTS : indexed_as

Loading

Note: edges.from_entity_id and edges.to_entity_id both reference entities.entity_id. Also, provenance.subject_id is polymorphic (entity_id or edge_id) based on subject_type, so that relation is enforced in application logic rather than a single SQL foreign key.

License

This project is licensed under the MIT License. See LICENCE.md.