Skip to content

v0.1.3

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 27 May 21:29
v0.1.3
131428f

ScholarOS Changelog

All notable features and capabilities of ScholarOS.

AI Copilot & Chat

  • AI-powered chat with multi-turn conversations across multiple concurrent chat tabs
  • Model selection per chat tab (OpenAI, Anthropic, Gemini, Ollama, OpenRouter, Vercel AI Gateway, OpenAI-compatible)
  • Web search toggle per message — enables real-time web search in the agent's context
  • Voice input — record audio transcribed via Deepgram speech-to-text
  • Text-to-speech — AI responses spoken aloud via ElevenLabs (summary or full response)
  • Caveman mode — ultra-compressed response style that cuts ~75% token usage while maintaining technical accuracy
  • Chat history — persisted runs with context menu to open/delete; clear-all option
  • Streaming responses with live tool call expandable cards
  • @file mentions in chat input to reference knowledge files
  • Suggested topics — view and launch AI-generated study leads
  • Permission system — approve/deny/always tool execution permissions per session
  • Human-in-the-loop — agent can ask the user clarifying questions mid-conversation

Knowledge & File Management

  • Obsidian-compatible markdown vault — full wiki-link support ([[Note Name]])
  • File tree sidebar with expand/collapse, rename, delete (to trash), copy path
  • Markdown editor (TipTap-based) with headings, lists, bold, italic, links, images, code blocks, tables, and frontmatter YAML
  • Inline content blocks — prompt blocks, Mermaid diagrams, charts, transcripts, embeds, iframes
  • Image upload to .assets/ folder within the vault
  • PDF viewer — inline PDF rendering in the app
  • HTML viewer — renders .html files inline
  • Generic file viewer — raw content display for any other file type
  • Office file handling — opens .docx, .xlsx, .pptx in the system default app
  • Graph view — force-directed visualization of wiki-linked knowledge files
  • Bases (spreadsheet view) — table view of notes with columns, filters, sort; special folders (People, Organizations, Projects, Topics) have preset views
  • Artifacts — browser for all non-markdown files (HTML, MD, JSON, CSV, TXT) in the vault
  • Version history — git-based file history panel: view past commits, restore previous versions
  • Full-text search — across knowledge files and chat history with type filters
  • Auto-save with "Saving..." / "Saved" status indicator
  • File tabs — multiple concurrent file views with tab switching (Cmd+1-9, Cmd+Shift+[/])
  • Back/forward navigation through view history

Vault System

  • Vault selection via native folder picker — choose any directory as your workspace
  • Vault switching at runtime — triggers watcher restart and full context reload
  • Workspace file watcher (chokidar-based) — live file change detection with debounced events
  • Hidden directories — agents/, bases/, config/, events/, logs/ are reserved for internal use

AI Skills (Loadable Agent Capabilities)

The AI copilot can dynamically load specialized skills to perform complex tasks:

  • Document Collaboration (doc-collab) — create, edit, refine notes; direct-edit and approval modes; wiki-linking
  • Organize Files (organize-files) — find, organize, tidy files; clean up Desktop/Downloads
  • PowerPoint Presentations (pptx) — create and edit slide decks with charts, shapes, images, tables, slide masters using PptxGenJS or XML manipulation
  • Word Documents (docx) — create and edit .docx files with full formatting, tables, images, tracked changes, comments, TOC, headers/footers, multi-column layouts
  • Excel Spreadsheets (xlsx) — create and edit .xlsx files with formulas, formatting, charts, financial model color standards, formula recalculation via LibreOffice
  • PDF Processing (pdf) — merge, split, rotate, extract text/tables, create new PDFs, OCR, encrypt/decrypt, add watermarks, fill forms, extract images
  • Web Artifacts Builder (web-artifacts-builder) — build self-contained React+TypeScript+Tailwind+shadcn/ui HTML files
  • Revision Guide (revision-guide) — generate comprehensive standalone HTML study guides with exam weight badges, diagrams, comparison tables, quick-fire checklists
  • YouTube Video Link Finder (youtube-video-workflow) — find real YouTube video links for topics via curl-based search scraping
  • Anki Flashcards (anki-flashcards) — create, manage, push Anki flashcards from course materials; Q&A and cloze-deletion cards synced via AnkiConnect
  • MCP Integration (mcp-integration) — discover, execute, and integrate tools from external MCP servers
  • Composio Integration (composio-integration) — interact with 25+ third-party services (Gmail, GitHub, Slack, LinkedIn, Notion, Jira, Google Sheets, etc.)
  • Deletion Guardrails (deletion-guardrails) — confirmation process before removing workflows, agents, or dependencies
  • App Navigation (app-navigation) — navigate the app UI, open notes, switch views, filter/search knowledge base, manage saved bases
  • Browser Control (browser-control) — control the embedded browser pane programmatically
  • Caveman Mode (caveman) — toggle compressed assistant tone (lite/full/ultra/wenyan variants)
  • Builtin Tools Reference (builtin-tools) — understand and use builtin tools in agent definitions

Builtin Copilot Tools

Tools available to the AI agent natively (no skill load required):

Workspace Operations

  • Read, write, edit, delete, rename, copy files and directories
  • Glob pattern search and regex content search (ripgrep)
  • File/directory existence checks, stats, recursive directory listing
  • Atomic writes with ETag conflict detection

File Parsing

  • Parse text from PDF, PPTX, DOCX, XLSX/CSV, and images (PNG/JPG)
  • Fallback chain: pdf-parse -> Tesseract.js OCR -> LLM vision
  • Image OCR and LLM vision parsing
  • File classification into course folders via local embeddings (zero API calls)

Execution

  • Shell command execution with abort support and signal handling (SIGTERM -> SIGKILL)

Web

  • Web search via embedded browser (opens Google, reads search results)

Memory

  • Save user observations/preferences to agent memory inbox

Integrations

OAuth Connected Accounts

  • Google — email, drive, calendar (via OAuth or Composio)
  • Microsoft — Outlook, Teams, OneDrive
  • Zoom — meeting records
  • Notion — workspace and database access
  • Slack — workspace integration with enable/disable and workspace listing
  • Gmail — direct email access

Composio Third-Party Toolkits (25 curated)

  • Communication: Gmail, Slack, Microsoft Outlook, Microsoft Teams
  • Productivity: Google Sheets, Google Docs, Notion, Asana, Trello, Airtable, Calendly, Cal.com
  • Development: GitHub, Jira, Linear
  • CRM: HubSpot, Salesforce
  • Social: LinkedIn, X (Twitter), Reddit
  • Storage: Google Drive, Dropbox, OneDrive
  • Support: Intercom, Zendesk

Each toolkit provides multiple tools (e.g., Gmail: fetch/send/trash/list labels; GitHub: list issues/create PR/repo operations).

MCP (Model Context Protocol)

  • STDIO transport — local command-based MCP servers
  • Streamable HTTP transport — HTTP-based MCP servers
  • SSE transport — Server-Sent Events fallback
  • Server connection management with per-server state tracking (connected/disconnected/error)
  • Paginated tool listing and tool execution
  • Configuration persistence in config/mcp.json

ScholarOS Account

  • OAuth 2.0 + PKCE authentication via ScholarOS website (Clerk-backed)
  • LLM proxy to OpenRouter
  • Subscription management with billing info retrieval
  • Token-based authentication with self-signed HMAC-SHA256 JWTs

File Export

Format Method
Markdown (.md) Direct file save
PDF HTML render -> Chromium printToPDF (A4, print backgrounds)
DOCX Markdown -> HTML -> html-to-docx
PPTX PptxGenJS (from scratch) or XML unpack/replace/pack (template editing)
XLSX openpyxl Python workflow with formulas, formatting, financial model color coding
HTML Self-contained React+Tailwind+shadcn/ui artifacts

Voice Features

  • Voice memo recording — record audio, auto-transcribe via Deepgram, save to Voice Memos/ folder
  • Text-to-speech — AI responses synthesized via ElevenLabs
  • Configurable — Deepgram and ElevenLabs API key settings

Ingestion & Processing

  • File ingestion — drag-drop or file picker to stage external files into the vault raw/ directory
  • Academic pipeline — PDF ingestion with metadata extraction (title, authors, abstract), course classification via local embeddings, contradiction detection, prerequisite graph construction, annotation processing

Knowledge Graph & Note Creation

  • Automated note creation agent — processes source files (emails, meeting transcripts, voice memos) into structured knowledge notes
  • Entity resolution — resolves people, organizations, projects to canonical names
  • State change detection — detects project status changes, open item resolution, role/title changes
  • Bidirectional wiki-linking enforced automatically (concept -> course AND course -> concept)
  • Configurable note types — Course Concept, Lecture Notes, Assignment, Paper Summary, Synthesis, Resource, Entity — each with custom folder, template, and extraction guide
  • Agent notes system — separate memory maintenance agent that processes emails, copilot conversations, and inbox entries; maintains user.md, preferences.md, style/email.md with timestamped bullet-point facts and automatic stale fact removal
  • Tag system — label-based filtering rules determining which emails warrant note creation
  • Knowledge index — scans all markdown files, builds categorized indexes for courses, concepts, lectures, assignments, papers, syntheses, resources, entities

Settings

  • Account — ScholarOS sign-in/out, billing info
  • Connected Accounts — manage OAuth connections (Google, Microsoft, Zoom, Notion, Slack, Gmail)
  • Models — LLM provider selection and configuration per category (chat, knowledge graph, meeting notes); hidden when signed into ScholarOS account
  • MCP Servers — JSON editor for config/mcp.json
  • Security — JSON editor for config/security.json (allowed shell commands)
  • Appearance — theme (Light / Paper / Dark / System), font style (Serif / Sans), font size (Small / Medium / Large)
  • Tools Library — Composio toolkit browser: search, connect/disconnect tool integrations

Onboarding (4 Steps)

  1. Vault selection (choose folder or demo vault)
  2. Theme preference (Light / Paper / Dark / System)
  3. LLM provider and model configuration
  4. Completion screen with "All set!"

Embedded Browser

  • Multi-tab browser within the app
  • Navigate, back, forward, reload
  • Tab management — create, switch, close tabs
  • Position/resize control, show/hide
  • Dangerous scheme blocking (javascript:, file:, chrome:)
  • State sync to renderer for live UI updates
  • Programmatic control via AI copilot (browser-control skill)

Agent Runs System

  • Full lifecycle management: create, stop (graceful then force), abort with SIGKILL/MCP force-close
  • Permission authorization flow (approve/deny/always)
  • Human-in-the-loop input requests
  • Run listing with pagination, deletion, clear-all
  • Message queue with event streaming (streaming LLM deltas, tool invocations, errors)
  • Sub-flow spawning capability

Keyboard Shortcuts

Shortcut Action
Cmd+K Open Command Palette (search + quick compose)
Cmd+L Toggle full-screen chat / restore two-pane view
Cmd+W Close active tab (file or chat)
Cmd+1 through Cmd+9 Switch to tab 1-9
Cmd+Shift+] Next tab
Cmd+Shift+[ Previous tab
Cmd+Z Undo (in markdown editor)
Cmd+Shift+Z / Cmd+Y Redo (in markdown editor)

System

  • Command Palette (Cmd+K) — unified search across knowledge files and chat history with quick AI message compose
  • Two-pane layout — center content + right chat sidebar; toggleable to full-screen chat
  • Sidebar toggle — show/hide left file tree
  • Electron version info — Chrome, Node.js, Electron versions
  • App restart capability
  • Slack workspace discovery via agent-slack CLI
  • Local development server for serving app content

Contributors

file
Assets 7
Loading