yar

KO | EN

gitlite — search

#typescript #ai-agents #ai #dsh-plugin #deepseek-harness #open-source #cli #claude-code #codex #developer-tools #react #windows

★ 10 Open GitHub ↗

An agentic chief-of-staff OS for executives, built on Claude Code

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Yar — AI Operating System for Executives

> Your strategic reference point.

Personal AI system for executives, built on [Claude Code](https://claude.ai/code). Transforms meeting transcripts into structured knowledge, manages tasks, tracks commitments, and provides daily briefings.

## What You Get

- **Single Inbox**: Drop anything in `inbox/` — the system auto-detects and routes it
- **Meeting Processing Pipeline**: Transcripts → speaker identification → task extraction → promise tracking
- **Task Management**: Inbox processing, priority management, weekly reviews (Getting Things Done methodology)
- **Daily Briefings**: Morning context with overdue items, upcoming deadlines, weekly goals
- **Self-Memory**: System learns your decisions, positions, preferences, and ideas over time
- **Promise Tracking**: Never forget what you promised or what was promised to you
- **People Cards**: Structured knowledge about everyone you work with
- **Decision & Ideas Log**: Immutable record of every decision and insight

## Quick Start

### Prerequisites

- [Claude Code](https://claude.ai/code) installed and authenticated
- macOS, Linux, or Windows (WSL)
- Meeting recorder that produces text transcripts (Plaud Note, Fireflies, Otter.ai, Zoom export, or manual notes)

### Setup (3 minutes)

1. Clone this repo:
   ```bash
   git clone https://github.com/kobelev-ai/yar.git my-assistant
   cd my-assistant
   ```

2. Run the setup wizard:
   ```bash
   claude
   # Then type: /setup
   ```

The wizard will:
   - Ask you 5-7 questions about your role, projects, and preferences
   - Generate your personalized workspace
   - Run a "brain dump" to capture your current tasks and priorities
   - Recommend filling the inbox with your documents
   - Commit everything to git (built-in undo)

3. (Optional) Set up automatic transcript fetcher:
   ```bash
   cp scripts/plaud_fetcher/.env.example scripts/plaud_fetcher/.env
   # Edit .env with your email credentials
   pip install python-dotenv
   ```

### Try the Demo First

Want to see how it works before setting up? Load the demo workspace:
```bash
claude
# Then type: /demo
```
This loads a realistic CEO workspace with projects, tasks, promises, and meeting history.

## Structure

```
inbox/                  # Drop ANYTHING here — system auto-processes
.claude/
  agents/               # AI agents (meeting-processor, todo-processor, etc.)
  skills/               # Slash commands (/setup, /brief, /status, /review)
  settings.json         # Claude Code settings
context/
  me.md                 # Your profile (generated by setup wizard)
  me/                   # Decisions, preferences, ideas, speaker patterns, boundaries
  projects/             # Active projects with people and context
  people/               # Key people cards
  meetings/             # Meeting transcripts (processed/)
  goals/                # Goals (quarterly, weekly)
  journal/              # Daily journal
tasks/
  todo.md               # Task list (live document)
  archive/              # Weekly archives
scripts/
  plaud_fetcher/        # Email-based transcript fetcher
templates/
  demo/                 # Demo data for showcasing the system
```

## Agents

| Agent | What it does |
|-------|-------------|
| `setup-wizard` | Initial workspace setup from intake questions + brain dump |
| `session-starter` | Morning briefing: tasks, overdue promises, project status |
| `meeting-processor` | Full pipeline: transcript → tasks, promises, decisions, people, insights |
| `todo-processor` | Inbox processing, task classification, priority management |

## Skills (Slash Commands)

| Command | Description |
|---------|-------------|
| `/setup` | Initial workspace setup (first time) |
| `/demo` | Load demo workspace with sample data |
| `/inbox` | Process everything in global inbox |
| `/brief <project>` | Quick project brief |
| `/status` | System overview — tasks, promises, projects |
| `/review` | Interactive weekly review |

## Daily Workflow

**Morning:**
1. Open Claude Code in the project folder
2. Session starter runs automatically with your briefing
3. Review priorities, pick focus tasks for the day

**During the day:**
- Drop anything into `inbox/` and say `/inbox`
- Just tell Claude a task, idea, or thought — it saves automatically

**End of day:**
- `/status` to see what's open
- Review completed tasks, update priorities

**Weekly:**
- `/review` for a full weekly review with archiving

## Key Concepts

### Single Inbox
Everything goes into one folder. Meeting transcripts, documents, notes, screenshots — just throw it in `inbox/`. The system detects what it is and routes it to the right place.

### Collector Behavior
The system proactively captures decisions, preferences, ideas, and people info from every conversation. You don't need to say "remember this" — it just does.

### Boundaries
You control what gets processed. Add rules to `context/me/boundaries.md` to exclude sensitive topics. The system never invents confidentiality rules on its own.

## Privacy

All data stays on your machine. No cloud sync, no external APIs (except Claude Code itself). Meeting transcripts, decisions, and personal data are stored as local files. Use `.gitignore` to exclude sensitive data from version control if pushing to a remote repo.

## Customization

The system is designed to be extended:
- Add new agents in `.claude/agents/`
- Add slash commands in `.claude/skills/`
- Modify `CLAUDE.md` to change behavior
- Add integrations in `scripts/`

## License

MIT

yar

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README