🚀 AgentScript

A Domain-Specific Language for AI-Powered Automation

AgentScript is a simple, powerful DSL that chains Gemini AI with Google APIs to automate complex workflows in just a few lines of code.

search "AI trends 2026" -> summarize -> translate "Spanish" -> email "[email protected]"

🎯 What is AgentScript?

AgentScript turns natural workflows into executable pipelines. Instead of writing hundreds of lines of code to:

Search the web
Summarize results with AI
Generate images
Create videos
Send emails

You write one line:

search "topic" -> summarize -> image_generate "visual" -> email "[email protected]"

🏆 Hackathon Submission

Google Gemini API Developer Competition

Key Features

34 Commands - Research, documents, multimedia, Google services
Parallel Execution - Run multiple tasks concurrently
Pipeline Chaining - Output of one command feeds into the next
Natural Language Mode - Describe what you want in plain English
Gemini Integration - Text, images, video, and speech generation

Google APIs Used

API	Commands
Gemini 2.5 Flash	`ask`, `summarize`, `analyze`, `translate`
Imagen 4	`image_generate`
Veo 3.1	`video_generate`, `images_to_video`
Gemini TTS	`text_to_speech`
Gmail	`email`
Calendar	`calendar`, `meet`
Drive	`drive_save`
Docs	`doc_create`
Sheets	`sheet_create`, `sheet_append`
Forms	`form_create`, `form_responses`
YouTube	`youtube_search`, `youtube_upload`, `youtube_shorts`
Tasks	`task`
People	`contact_find`

⚡ Quick Start

1. Prerequisites

# Install Go 1.22+
brew install go  # macOS
# or download from https://go.dev

# Install ffmpeg (for video/audio)
brew install ffmpeg  # macOS
sudo apt install ffmpeg  # Ubuntu

2. Setup

# Clone the repository
git clone https://github.com/vinodhalaharvi/agentscript.git
cd agentscript

# Copy environment template
cp .env.example .env

# Add your API keys to .env
GEMINI_API_KEY=your_gemini_api_key
GOOGLE_CREDENTIALS_FILE=credentials.json
GOOGLE_TOKEN_FILE=token.json

3. Google OAuth Setup

Go to Google Cloud Console
Create a project and enable APIs:
- Gmail API
- Google Calendar API
- Google Drive API
- Google Docs API
- Google Sheets API
- Google Forms API
- YouTube Data API v3
Create OAuth 2.0 credentials (Desktop app)
Download as credentials.json

4. Build & Run

# Build
make build

# Run a simple command
make run EXPR='ask "Hello, what can you do?"'

# Run a script file
make run-file FILE=examples/simple-research.as

# Interactive natural language mode
make repl

📖 Language Syntax

Basic Pipeline

command "arg" -> command "arg" -> command "arg"

Parallel Execution

parallel {
    search "topic 1" -> summarize
    search "topic 2" -> summarize
    image_generate "visual"
}
-> merge
-> email "[email protected]"

Comments

// This is a comment
search "query" -> summarize

🛠 All 34 Commands

Core Commands

Command	Description	Example
`search "query"`	Web search via Gemini	`search "AI news"`
`ask "prompt"`	Ask Gemini anything	`ask "Explain quantum computing"`
`summarize`	Summarize piped content	`search "topic" -> summarize`
`analyze`	Analyze piped content	`read "data.csv" -> analyze`
`save "file"`	Save to file	`-> save "output.txt"`
`read "file"`	Read file contents	`read "input.txt" -> summarize`
`stdin "prompt"`	Read user input	`stdin "Enter topic: " -> search`
`translate "lang"`	Translate text	`-> translate "Japanese"`

Google Workspace

Command	Description	Example
`email "to"`	Send email	`-> email "[email protected]"`
`calendar "event"`	Create calendar event	`-> calendar "Meeting tomorrow 2pm"`
`meet "event"`	Create event with Meet link	`-> meet "Team sync"`
`drive_save "path"`	Save to Google Drive	`-> drive_save "reports/q1"`
`doc_create "title"`	Create Google Doc	`-> doc_create "Report"`
`sheet_create "title"`	Create spreadsheet	`-> sheet_create "Data"`
`sheet_append "id"`	Append to sheet	`-> sheet_append "sheet_id"`
`task "title"`	Create Google Task	`-> task "Follow up"`
`contact_find "name"`	Find contact	`contact_find "John"`
`form_create "title"`	Create Google Form	`-> form_create "Survey"`
`form_responses "id"`	Get form responses	`form_responses "form_id"`

YouTube

Command	Description	Example
`youtube_search "query"`	Search YouTube	`youtube_search "Go tutorials"`
`youtube_upload "title"`	Upload video	`-> youtube_upload "My Video"`
`youtube_shorts "title"`	Upload as Short	`-> youtube_shorts "Quick Tip"`

Multimedia (Gemini)

Command	Description	Example
`image_generate "prompt"`	Generate image (Imagen 4)	`image_generate "sunset"`
`image_analyze "path"`	Analyze image	`image_analyze "photo.jpg"`
`video_generate "prompt"`	Generate video (Veo 3.1)	`video_generate "ocean waves"`
`video_analyze "path"`	Analyze video	`video_analyze "clip.mp4"`
`images_to_video "paths"`	Images to video	`-> images_to_video`
`text_to_speech "voice"`	Convert text to speech	`-> text_to_speech "Kore"`
`audio_video_merge "out"`	Merge audio + video	`-> audio_video_merge "final.mp4"`
`image_audio_merge "out"`	Image + audio to video	`-> image_audio_merge "video.mp4"`

Travel & Places

Command	Description	Example
`places_search "query"`	Search for places	`places_search "cafes Tokyo"`
`maps_trip "name"`	Create trip map URL	`-> maps_trip "Tokyo Trip"`

Control

Command	Description	Example
`merge`	Merge parallel outputs	`parallel { ... } -> merge`
`list "path"`	List directory	`list "."`

🎬 Example Workflows

1. Research Report

make run EXPR='search "renewable energy 2026" -> summarize -> doc_create "Energy Report" -> email "[email protected]"'

2. Travel Planner

make run-file FILE=examples/travel-planner.as

parallel {
    places_search "attractions Dubrovnik"
    places_search "restaurants Dubrovnik"
    places_search "hotels Dubrovnik"
}
-> merge
-> ask "Create 5-day itinerary"
-> maps_trip "Dubrovnik Adventure"
-> translate "Croatian"
-> email "[email protected]"

3. AI News Video (2 minutes)

make run-file FILE=examples/news-2min.as

parallel {
    search "top US news today"
    -> ask "Write 2-minute news script"
    -> text_to_speech "Charon"
    -> save "narration.wav"

    image_generate "news studio background" -> save "bg.png"
}
-> merge
-> image_audio_merge "news.mp4"
-> youtube_upload "Daily News Update"

4. Multilingual Content

make run EXPR='ask "Write a welcome message" -> translate "Spanish" -> translate "Japanese" -> translate "French" -> save "translations.txt"'

5. Event Planning with RSVP

make run EXPR='ask "Plan a team offsite agenda" -> form_create "Offsite RSVP" -> calendar "Team Offsite next Friday 9am" -> email "[email protected]"'

6. Competitor Analysis

make run-file FILE=examples/competitor-analysis.as

7. Interactive Mode

make repl
> search for latest AI research and summarize the top findings
> create an image of a futuristic city and save it
> translate "hello world" to 5 languages

📁 Project Structure

agentscript/
├── main.go           # Entry point
├── grammar.go        # DSL parser (Participle)
├── runtime.go        # Command execution engine
├── client.go         # Gemini API client
├── google.go         # Google Workspace APIs
├── github.go         # GitHub Pages deployment
├── translator.go     # Natural language to DSL
├── claude.go         # Claude API (optional)
├── examples/         # Example scripts
│   ├── mega-showcase.as      # 100-line comprehensive demo
│   ├── travel-planner.as     # Travel planning workflow
│   ├── news-2min.as          # 2-minute news video
│   ├── youtube-shorts.as     # YouTube Shorts creation
│   └── ...
├── Makefile          # Build commands
├── .env.example      # Environment template
└── README.md         # This file

🔧 Makefile Commands

make build                    # Build binary
make run EXPR='...'          # Run inline expression
make run-file FILE=path.as   # Run script file  
make repl                    # Interactive mode
make test                    # Run tests
make clean                   # Clean build artifacts

🎤 TTS Voices

Available voices for text_to_speech:

Kore - Female, warm
Charon - Male, deep (great for news)
Puck - Male, energetic
Aoede - Female, calm

ask "Hello world" -> text_to_speech "Kore" -> save "greeting.wav"

🌐 Environment Variables

# Required
GEMINI_API_KEY=your_gemini_api_key

# For Google Workspace integration
GOOGLE_CREDENTIALS_FILE=credentials.json
GOOGLE_TOKEN_FILE=token.json

# Optional - for GitHub Pages deployment
GITHUB_CLIENT_ID=your_client_id
GITHUB_CLIENT_SECRET=your_client_secret

# Optional - for Claude as alternative LLM
CLAUDE_API_KEY=your_claude_key

🚨 Troubleshooting

"GEMINI_API_KEY not set"

export GEMINI_API_KEY=your_key
# or add to .env file

"ffmpeg not found"

brew install ffmpeg  # macOS
sudo apt install ffmpeg  # Ubuntu

"OAuth scope insufficient"

rm token.json  # Delete old token
make run EXPR='...'  # Re-authenticate with new scopes

"Veo quota exceeded"

Use image_audio_merge instead of images_to_video
Or wait for daily quota reset
Or request quota increase in Google Cloud Console

📊 Architecture

┌─────────────────────────────────────────────────────────────┐
│                      AgentScript                            │
├─────────────────────────────────────────────────────────────┤
│  Natural Language ──► Translator ──► DSL Parser            │
│         │                               │                   │
│         ▼                               ▼                   │
│  "summarize AI news"        search "AI" -> summarize       │
├─────────────────────────────────────────────────────────────┤
│                     Runtime Engine                          │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐       │
│  │ Gemini  │  │ Google  │  │ GitHub  │  │ ffmpeg  │       │
│  │  APIs   │  │  APIs   │  │  API    │  │         │       │
│  └─────────┘  └─────────┘  └─────────┘  └─────────┘       │
│       │            │            │            │              │
│       ▼            ▼            ▼            ▼              │
│  Text/Image    Gmail/Cal     Pages      Audio/Video        │
│  Video/TTS     Drive/Docs    Deploy      Processing        │
└─────────────────────────────────────────────────────────────┘

🏅 Why AgentScript?

Traditional Approach	AgentScript
500+ lines of Python	5 lines of DSL
Multiple API libraries	One unified syntax
Complex async handling	Built-in parallelism
Manual error handling	Automatic retries
Separate scripts	Chainable pipelines

📜 License

MIT License - See LICENSE file

👨‍💻 Author

Vinod Halaharvi

GitHub: @vinodhalaharvi
Email: [email protected]

🙏 Acknowledgments

Google Gemini API Team
Anthropic Claude (alternative LLM support)
Participle Parser Library

Built for the Google Gemini API Developer Competition 2026 🏆

Name		Name	Last commit message	Last commit date
Latest commit History 30 Commits
examples		examples
.env.example		.env.example
.gitignore		.gitignore
Makefile		Makefile
README.md		README.md
claude.go		claude.go
client.go		client.go
github.go		github.go
go.mod		go.mod
go.sum		go.sum
google.go		google.go
grammar.go		grammar.go
main.go		main.go
runtime.go		runtime.go
translator.go		translator.go

Folders and files

Latest commit

History

Repository files navigation

🚀 AgentScript

🎯 What is AgentScript?

🏆 Hackathon Submission

Key Features

Google APIs Used

⚡ Quick Start

1. Prerequisites

2. Setup

3. Google OAuth Setup

4. Build & Run

📖 Language Syntax

Basic Pipeline

Parallel Execution

Comments

🛠 All 34 Commands

Core Commands

Google Workspace

YouTube

Multimedia (Gemini)

Travel & Places

Control

🎬 Example Workflows

1. Research Report

2. Travel Planner

3. AI News Video (2 minutes)

4. Multilingual Content

5. Event Planning with RSVP

6. Competitor Analysis

7. Interactive Mode

📁 Project Structure

🔧 Makefile Commands

🎤 TTS Voices

🌐 Environment Variables

🚨 Troubleshooting

"GEMINI_API_KEY not set"

"ffmpeg not found"

"OAuth scope insufficient"

"Veo quota exceeded"

📊 Architecture

🏅 Why AgentScript?

📜 License

👨‍💻 Author

🙏 Acknowledgments

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages