Mega Agenda

A personal productivity powerhouse that lives in your system tray. Task management, AI chat, goal research, bank account sync, Twitter posting, RSS feeds, embedded terminal, voice commands, and more — all in a single compact Electron window.

Built with Electron + React + TypeScript + Tailwind CSS + Claude AI.

Features

Task Management

Organize tasks by category (Work, Health, Finance, etc.) with priorities, due dates, and color-coded icons
Recurring tasks (daily, weekly, monthly) with auto-reset
Category cards with completion rings and quick filters
Add custom categories with emoji icons and colors

Dashboard

AI-generated morning briefing (overdue tasks, priorities, streak info)
Progress ring with motivational messages
90-day activity heatmap
Live debt summary card (auto-syncs from connected bank accounts)
Weekly stats and streak tracking
Category grid for quick navigation

Daily Journal & Calendar

Date-based note editor with auto-save
Calendar picker with visual indicators for days with notes
Navigate between days or jump to today
AI-powered weekly reviews with task counts, focus time, and category breakdowns
Historical review archive
Routines engine — scheduled automations that run on cadences (app launch, daily, weekly, interval)
- 5 routine types: morning briefing, PR monitor, email digest, weekly review, custom command
- Results surface as cards in calendar day panel with expandable markdown content
- Native OS notifications on routine completion

Bank Accounts (Live Sync)

SimpleFIN Bridge integration — 16,000+ banks via MX aggregator ($15/year, read-only)
Teller integration — direct bank connections (free, real-time)
Connect wizard: choose provider, paste token, auto-sync
Live debt/balance tracking across all connected accounts
Financial overview: total debt, total assets, per-account breakdown
Transaction list with account filtering and pending indicators
Connection status monitoring with manual sync
Auto-sync on dashboard load when stale (>1 hour)
MD5-based transaction deduplication

Feed Reader (RSS + Twitter)

Multi-section feed aggregation (AI/LLMs, World News)
Add/remove RSS feeds with 30+ suggested sources (HN, Reddit, arXiv, etc.)
Twitter/X list integration via bearer token
AI-powered article summarization per section
Cached summaries with refresh

Content Writer & Tweet Scoring

Multi-format content creation: tweets, threads, blog posts, articles, Discord posts, newsletters
AI-powered research → draft → refine pipeline with streaming output
Draft management with status tracking (researching → drafting → refining → ready)
Quick actions: Provocative, Flip It, ELI5, Pain Point, Mic Drop, Story (tweets) or Shorter, Punchier, Hook, Technical, Simpler, Contrarian (other formats)
Tweet Scoring Framework — Hook/Clarity/Viral scoring (1-10) with calibration anchors
- Qualitative feedback per tweet: strengths, weaknesses, improvement suggestions
- Score history with daily snapshots and 7-day running averages
- LLM-powered pattern extraction from scored tweets (positive/negative patterns)
- Intuition builder: injects top/bottom performing tweets into generation prompts
- Auto-refinement: weak tweets (avg < 6) get automatic refinement pass using score feedback
Import from Sessions — pull AI-generated tweet insights from AgentsView and import as content drafts for scoring
Copy-to-clipboard per tweet with character counter

Twitter / Social Media Manager

Compose single tweets or threaded posts
Draft management with status tracking (draft → refining → ready → posted)
AI writing assistant: brainstorm, refine, analyze tweet effectiveness
Pre-built refinement actions (make punchier, add hook, contrarian angle, etc.)
Persona system — 5 built-in (Paul Graham, Naval, Snarky Critic, Builder, Thought Leader) + custom personas
OAuth v2 integration for posting directly to Twitter
Sequential thread posting with reply chaining
Character counter and copy-to-clipboard

Chat (Claude AI)

Multiple conversation threads with persistent history
Streaming responses with abort support
Multi-provider LLM support (Claude, Gemini, Groq, OpenRouter)
Model selection (Sonnet 4.5, Haiku 4.5, Opus 4.6)
System prompt modes: default, context-aware (includes tasks/streak/notes), or custom
Token usage tracking per message
Memory integration — extract and inject memories into context
Smart Query (/ask) — RAG-powered Q&A with semantic search over knowledge base

Code Terminal

Embedded terminal powered by node-pty + xterm.js
Full Windows CMD shell with copy/paste (Ctrl+C/V)
Auto-fit on window resize
Custom dark color theme
Always-mounted for instant access

AI Tasks (Kanban Board)

4-column workflow: Backlog → Todo → In Progress → Done
Tasks with title, description, priority, and tags
Auto-launch Claude Code CLI sessions from tasks
GitHub repo search for context injection
Opens in external terminal windows
Inline editing with expandable detail view

AI Agent Orchestration

Named agents with roles (Engineer, Researcher, Reviewer, PM), task types, and working directories
Heartbeat scheduling — interval or daily triggers that auto-launch Claude Code sessions
Issue tracking Kanban — Backlog → Todo → In Progress → Review → Done, assigned to agents
Auto-complete — detects finished Claude Code sessions via JSONL mtime, no manual "Mark Done"
Budget management — monthly budgets per agent, auto-pause at 80% threshold with desktop notification
Retry with backoff — failed terminal launches retry 3x (30s/60s/120s) before marking failed
Cost dashboard — total runs, success rate, total spend, avg duration across agents
Export — CSV/JSON export of agents, issues, runs, and cost events
Run auto-tagging — generates tags from agent role, task type, status, priority on completion
Dashboard stats with 4 summary cards in Overview

Sessions (AgentsView Analytics)

Full analytics dashboard for Claude Code session history, powered by AgentsView
Overview — summary cards (sessions, messages, projects, active days), CSS-grid activity heatmap, project breakdown, top sessions
Tools — horizontal bar charts showing tool usage by category with weekly trend bars
Velocity — p50/p90 metrics for turn cycle time, first response, messages/min, chars/min, tool calls/min; complexity breakdown
Sessions — SVG histograms for session length, duration, and autonomy distributions; recent sessions table
Insights — sync status card, daily AI-generated insight cards from AgentsView
Online/offline detection with graceful degradation

Roadmap / Goal Planning

Create goals with category, priority, status, target quarter/year
AI-generated research questions and guidance needs per goal
Parallel batch research (3 topics at a time, CLI-first with API fallback)
Individual topic reports saved as markdown
Action plan synthesis from all research
Context questionnaire for personalization
Goal workspaces with deliverables and git tracking

Master Plan

Cross-goal master plan synthesis using all research + context files
Hybrid RAG-powered smart context retrieval — vector semantic search + BM25 full-text search, merged via Reciprocal Rank Fusion
Local embeddings (MiniLM-L6-v2, 384-dim) with LanceDB vector index (~13,400+ chunks)
BM25 full-text index (MiniSearch) with fuzzy matching and prefix search
Claude Code session indexing — 968+ past sessions parsed and searchable alongside memory files
Falls back to vector-only or BM25-only if either index is unavailable
Task extraction from plan (10-30 actionable items)
Execution dashboard with task launching via Claude Code CLI
Auto-polling to match launched tasks with CLI sessions

Context Files (Knowledge Base)

File manager for ~/.claude/memory/ directory
Create, edit, delete files and folders
Upload files from disk
Domain scaffolding — auto-generates 8 domain folders (career, health, financial, relationships, learning, projects, personal, creative) with profile templates
Auto-synced goal context files (_overview.md, topic reports, action plans)
AI-driven reorganization tool — analyzes files, suggests moves/merges/deletes, backs up before executing

Memories (Structured Knowledge)

Create, edit, delete, archive, and pin memories
Importance levels (1-3) with topic/tag coloring
Source tracking: chat, CLI session, journal, task, AI task, or manual
AI-powered batch extraction from all sources
Topic management: rename, recolor, merge, delete
Search by title/content/topic, filter by source type
Memory injection into chat context
Knowledge pack compression — cluster and summarize memories into dense packs
Memory health monitoring with auto-pruning

In-App Guide Agent

Floating help widget (bottom-right FAB) available on every tab
Chat panel powered by fast-tier LLM with comprehensive system prompt
Covers all 20 tabs, key workflows, keyboard shortcuts, setup steps, and architecture
Conversation resets on close for lightweight context

Lab (Experimental Tools)

Single-file compression testing
Embedding similarity testing
Compression audit with coverage scoring

CLI History Viewer

Browse all Claude Code CLI sessions from ~/.claude/projects/
View full conversations with pagination (50 messages at a time)
Full-text search across all sessions
Extract memories from sessions
Quick-link from AI tasks to related sessions

Voice Commands (Experimental)

Currently being reworked — migrating from Web Speech API (not supported in Electron 28) to local Whisper transcription
Local fallback parsing for navigation and basic commands
AI parsing via Claude for complex commands
Visual feedback: pulsing indicator, transcript banner, success/error states

Command	Example
Switch tab	"Go to feed", "Show dashboard"
Add task	"Add task to work: finish the report, high priority"
Complete task	"Mark done buy groceries"
New task modal	"New task"
Journal note	"Note: met with Sarah about the project"
Summarize feed	"Summarize AI news"

Pomodoro Timer

25-minute focus sessions linked to tasks
Task picker with search
Auto-start breaks (5 min short / 15 min long every 4 sessions)
Native OS notifications on session complete
Session counter for the day

Settings

Multi-provider LLM configuration (Claude, Gemini, Groq, OpenRouter)
API key storage and verification per provider
Twitter OAuth credentials (API keys, access tokens)
Category management (add/delete with icons and colors)
RSS feed management with suggested feeds
Keyboard shortcuts reference

Keyboard Shortcuts

Key	Action
N	New task
D	Dashboard
T	Tasks
L	List view
J	Journal
B	Bank accounts
F	Feed
M	Social / Twitter
H	Chat
C	Code terminal
A	AI Tasks
G	Agents
Y	Context files
R	Roadmap
S	Settings
V	Voice command
P	Pomodoro timer
1-7	Select category (on dashboard)
Esc	Back / Close

Architecture

Tech Stack

Frontend: React 18, TypeScript, Tailwind CSS, Vite
Backend: Electron 28 (main + preload + renderer)
State: Zustand stores (app, task, chat, social, content, agent, sessions)
Database: JSON file (mega-agenda.json) persisted to %APPDATA%/mega-agenda/; reads AgentsView SQLite for session analytics
AI: Multi-provider LLM support — Claude API (Sonnet 4.5 / Haiku 4.5 / Opus 4.6), Gemini, Groq, OpenRouter
Embeddings: @xenova/transformers with MiniLM-L6-v2 (384-dim, ~22MB, cached locally)
Search: Hybrid — LanceDB vector + MiniSearch BM25, merged via Reciprocal Rank Fusion (RRF, k=60)
Terminal: node-pty + xterm.js
Speech: Local Whisper via @xenova/transformers (migrating from Web Speech API, which is unsupported in Electron 28)
Twitter: Twitter API v2 (OAuth 1.0a for posting, bearer token for reading)
Bank Sync: SimpleFIN Bridge (16K+ banks via MX) + Teller (direct connections)
AgentsView: Session analytics via HTTP proxy to AgentsView (localhost:8090) + direct SQLite reads for insights

Data Storage

JSON DB: %APPDATA%/mega-agenda/mega-agenda.json — tasks, categories, notes, stats, settings, conversations, drafts, goals, memories, bank connections, bank accounts, transactions, agents, issues, heartbeat runs, cost events, routines, tweet patterns, score snapshots
Context files: ~/.claude/memory/ — markdown knowledge base with domain folders and goal research
Master plans: ~/.claude/master-plans/ — dated plan files
Vector DB: %APPDATA%/mega-agenda/vector-db/ — LanceDB vector index for semantic search
BM25 index: %APPDATA%/mega-agenda/bm25-index.json — MiniSearch full-text index
Hash sidecars: %APPDATA%/mega-agenda/vector-db-hashes.json + vector-db-session-hashes.json — incremental indexing
Model cache: %APPDATA%/mega-agenda/models/ — cached MiniLM and Whisper models

Project Structure

electron/                  # Main process
  main.ts                  # Entry point, BrowserWindow, model loading
  preload.ts               # contextBridge API (window.electronAPI)
  database.ts              # JSON DB with all CRUD operations
  llm.ts                   # Multi-provider LLM (Claude, Gemini, Groq, OpenRouter)
  research.ts              # Goal research, master plan, RAG
  embeddings.ts            # MiniLM-L6-v2 local embeddings
  whisper.ts               # Local Whisper transcription
  vector-store.ts          # Hybrid search: LanceDB vector + BM25, RRF fusion
  bm25-index.ts            # MiniSearch BM25 full-text index with disk persistence
  session-parser.ts        # Claude Code JSONL session parser → searchable chunks
  knowledge-pack.ts        # Knowledge compression and clustering
  memory.ts                # Memory extraction from sources
  smart-query.ts           # RAG-powered Q&A streaming (hybrid search)
  content-writer.ts        # Content draft generation, scoring, pattern extraction
  agents.ts                # Agent orchestration: heartbeat scheduler, session polling, cost aggregation
  bank-sync/               # Bank account sync
    simplefin.ts           # SimpleFIN API client
    teller.ts              # Teller API client
    sync.ts                # Sync orchestrator with deduplication
  ipc/                     # IPC handler modules (19 modules)
    index.ts               # Handler registration barrel
    tasks.ts, notes.ts, chat.ts, twitter.ts, rss.ts, ai.ts,
    memory.ts, system.ts, knowledge-pack.ts, bank-sync.ts,
    content.ts, agents.ts, calendar.ts, routines.ts,
    network.ts, social.ts, outreach.ts, guide-chat.ts

src/                       # Renderer (React)
  App.tsx                  # Root component
  types/index.ts           # Shared types + ElectronAPI interface
  store/                   # Zustand stores (app, task, chat, social, content, agent, sessions)
  hooks/                   # Custom hooks (9 files)
  utils/                   # Utilities (cn, formatting, markdown, sounds, tts)
  components/
    layout/                # AppShell, TitleBar, TabNavigation, ContentArea
    agents/                # AgentsTab, AgentCard, AgentForm, IssueBoard, CostDashboard, HeartbeatHistory
    bank-sync/             # AccountsTab, ConnectBankDialog, DebtSummary, ConnectionStatus
    chat/                  # ChatView, ChatSidebar, ChatMessages, ChatInput, SmartQuery
    roadmap/               # GoalForm, GoalDetail, Timeline, TopicList, Execution, MasterPlan
    sessions/              # SessionsTab, Overview, Tools, Velocity, Sessions, Insights sub-views
    social/                # TweetEditor, ThreadEditor, DraftSelector, PersonaSelector, AIAssist
    settings/              # Settings, AIProvider, Categories, Feeds, Twitter
    ui/                    # Primitives (Button, Input, Card, Badge, Dialog, etc.)
    ...                    # Dashboard, Tasks, Feed, Terminal, Memories, ContentTab, GuideAgent, Lab, etc.

Window Management

Frameless window with custom title bar
System tray icon with context menu (Open, Quick Add Task, Quit)
Single instance lock — second launch focuses existing window
Hide on close (minimize to tray)

Getting Started

Prerequisites

Node.js 18+
npm

Install & Run

git clone https://github.com/YOUR_USERNAME/mega-agenda.git
cd mega-agenda
npm install
npm run dev

Build Portable EXE

npm run build

The executable will be in release/.

Optional Setup

Claude API Key — Required for AI features (chat, research, summaries, master plan). Set in Settings tab.
Multi-LLM Providers — Optionally configure Gemini, Groq, or OpenRouter as alternative providers in Settings.
Twitter Credentials — Required for tweet posting. Set bearer token for reading, OAuth for posting in Settings.
Claude Code CLI — Install globally for CLI-first research and task launching: npm install -g @anthropic-ai/claude-code
SimpleFIN Bridge — $15/year for live bank sync. Sign up at simplefin.org, connect banks, paste setup token in Accounts tab.
AgentsView — Optional. Install for session analytics in the Sessions tab. Runs at localhost:8090.

Security

Content Security Policy (CSP) headers — strict in production, relaxed for Vite HMR in dev
Shell command injection protection — CMD metacharacter escaping in terminal launchers, validated URLs for shell.openExternal (http/https/mailto only)
HTML entity escaping in markdown renderer (XSS prevention)
Path traversal checks on context file IPC handlers
ASAR packaging enabled for production builds
No hardcoded PII — all personal references genericized

Name		Name	Last commit message	Last commit date
.claude		.claude
.github/workflows		.github/workflows
docs		docs
electron		electron
public		public
python		python
scripts		scripts
src		src
.eslintrc.cjs		.eslintrc.cjs
.gitignore		.gitignore
.prettierrc		.prettierrc
CLAUDE.md		CLAUDE.md
README.md		README.md
index.html		index.html
package-lock.json		package-lock.json
package.json		package.json
populate-claude-goal.js		populate-claude-goal.js
postcss.config.cjs		postcss.config.cjs
restore-baby-data.js		restore-baby-data.js
save-claude-research.js		save-claude-research.js
tailwind.config.cjs		tailwind.config.cjs
tsconfig.json		tsconfig.json
tsconfig.node.json		tsconfig.node.json
vite.config.ts		vite.config.ts
vitest.config.ts		vitest.config.ts

Latest commit

History

Folders and files

Repository files navigation