ERA 🚀

ERA of Replicating Agents - Create AI agents with custom prompts using Deno.

era-advertisement.mp4

Try us now!

https://agfactory-web.fly.dev/

🌟 Self-Improving Meta-Programming System

ERA doesn't just generate code—it improves itself. Every agent you create becomes a reusable utility for future agents. The system:

✨ Self-replicates: Generated agents can be promoted to utilities and injected into new agents
🔄 Self-corrects: FBI Director reviews, refines, and iterates on generated code automatically
🧬 Compounding growth: Each new agent expands the capabilities of the entire system

Features

🤖 CLI Tool - Create and manage AI agents with custom prompts
🧠 FBI System - AI-powered agent orchestration with prompt improvement & iteration
🏗️ Utility Promotion - AI converts agents to injectable utilities with examples.ts
🌐 Web Server - Hono-based server with Alpine.js frontend
☁️ Integrations - Daytona sandbox, Wandb AI, and Weave tracing

📢 Note: This project generates its own code organized as core/ (orchestration system), utils/ (reusable utilities used by the app itself for code generation, AI interactions, and integrations), and agents/ (experimental AI agents with specific tasks)
See docs/ARCHITECTURE.md for system design, docs/WORKFLOWS.md for usage patterns.

🚀 Quick Start

Prerequisites

Install Deno (if not already installed):

# macOS/Linux
curl -fsSL https://deno.land/install.sh | sh

# Windows
irm https://deno.land/install.ps1 | iex

Installation

# Clone the repository
git clone <your-repo-url>
cd ERA

# No installation needed! Deno downloads dependencies automatically.

📦 Project Structure

ERA/
├── core/                    # 🧠 Core FBI Orchestration System
│   ├── fbi.ts              # Main orchestrator
│   ├── director.ts         # AI decision maker & prompt improvement
│   ├── prep.ts             # File preparation
│   └── test-prep-integration.ts
├── utils/                   # 🔧 Utility Functions
│   ├── codegen/            # AI code generation
│   ├── daytona/            # Sandbox execution
│   ├── wandb/              # AI API wrapper
│   └── weave/              # Tracing
├── tests/                   # 🧪 Test Files
├── frontend/                # 🎨 UI Assets
│   └── index.html          # Alpine.js frontend
├── agents/                  # 🤖 Generated Agents
│   └── <agent-name>/
│       ├── index.ts         # Latest/best version
│       ├── agent.json       # Full metadata & history
│       └── iterations/      # Timestamped snapshots
├── docs/                    # 📚 Documentation
│   ├── ARCHITECTURE.md     # System design
│   ├── WORKFLOWS.md        # Usage patterns
│   ├── UTILITY-PROMOTION.md # Promotion system
│   └── NPM-PACKAGES-GUIDE.md # Package usage
├── cli.ts                   # 💻 CLI Interface
├── main.ts                  # 🌐 Hono Web Server
├── history.ts               # 📝 Type Definitions
├── deno.json               # Deno configuration & tasks
└── README.md

Agent Creation Process (How it works)

🔧 Usage

CLI Tool - Create Agents

Interactive Mode (Recommended) 🆕

deno task cli

This starts an interactive prompt with Quick Start Templates:

Choose a template or define your own:
- 🔢 FizzBuzz Solver - Simple code generation demo
- 🎭 Joke Generator - WandbChat + Weave tracing demo
- 🔍 Web Search Agent - Tavily search demo
- 🌐 AI Web Browser - Browserbase/Stagehand AI browsing demo
- 🤖 Multi-Agent System - Mastra framework with agents & workflows
- ✨ Define Your Own - Custom agent with AI-suggested name
Generate code using FBI Director + AI code generation
Choose location - agents/ (experimental) or utils/ (stable)
Refine (optional) - improve with natural language feedback
Promote (optional) - convert agent to reusable utility

New: Interactive Refinement 💡
After successful generation, you can refine the code by describing improvements:

"Add error handling"
"Make it faster with async/await"
"Add more test cases"
"Simplify the logic"

Each refinement creates a new iteration while preserving history.

New: Agent Continuation 🔄
Using an existing agent name automatically continues development:

Loads previous code and history
Builds upon existing functionality
Preserves all previous attempts
Tracks sessions across CLI invocations

Self-Improving Example 🌟

# 1. Create a joke-telling agent
deno task cli:create jokemeister --prompt "tell jokes with tracing"

# 2. Create an agent that USES the first agent
deno task cli:create joke-rater --prompt "rate jokes from jokemeister"

# 3. Promote to reusable utility (AI generates injection code)
deno task cli:promote jokemeister --dry-run  # Preview
deno task cli:promote jokemeister            # Execute

# 4. Now jokemeister is injectable into ALL future agents!
# The system grew its own capabilities 🚀

Quick Start

Try these pre-configured examples:

# FizzBuzz Solver (simple demo)
deno task start:fizzbuzz

# Joke Generator (WandbChat + Weave demo)
deno task start:jokemeister

# Web Search Agent (Tavily demo)
deno task start:tavily

# AI Web Browser (Browserbase/Stagehand demo)
deno task start:browserbase

# Multi-Agent System (Mastra framework demo)
deno task start:mastra

Command Line Mode

# Create an agent with a prompt
deno task cli:create my-agent --prompt "You are a helpful coding assistant"

# Or use the full command
deno run --allow-read --allow-write --allow-env cli.ts create my-agent -p "You are helpful"

Generated Files Structure

When you create an agent, the following structure is created:

agents/my-agent/
├── index.ts                    # Latest/best version (run this!)
├── agent.json                  # Full metadata with ALL attempts
└── iterations/                 # Historical snapshots
    ├── iteration-1-1760237068655.ts
    ├── iteration-2-1760237123456.ts
    └── iteration-3-1760237178901.ts

index.ts - Always contains the latest/best version to run
agent.json - Complete history of all attempts, prompts, execution results, and sessions (CLI invocations)
iterations/ - Timestamped snapshots so you can review or rollback

Sessions track each time you continue the agent with a new prompt, making it easy to see the development timeline.

Running Your Agents

After creating an agent, run it with:

deno run agents/my-agent/index.ts

Each agent is a self-contained TypeScript file that you can customize and extend.

Web Server - Hono Application

Development Mode (with hot reload)

deno task dev

This starts the Hono server at http://localhost:8000 with watch mode enabled.

Production Server

deno task serve

Runs the server on port 9995 with all necessary permissions.

Deploy to Deno Deploy

deno task prod

Backend Modules

Test Daytona Integration

deno task test:daytona

Runs Daytona sandbox tests for code execution.

Test Wandb Integration

deno task test:wandb

Tests Wandb Inference API with Weave tracing.

Test Weave Integration

deno task test:weave

Tests Weave tracing functionality.

⚙️ Configuration

Environment Variables

Create a .env file in the root directory:

# Daytona Configuration
DAYTONA_API_KEY=your_api_key
DAYTONA_API_URL=https://api.daytona.io

# Wandb Configuration
WANDB_API_KEY=your_wandb_key
WANDB_PROJECT=your-team/your-project

# AI Model Configuration
# Default model for Director, FBI, and Code Generation
AI_MODEL=Qwen/Qwen3-Coder-480B-A35B-Instruct

# Optional: Override specific models for different components
# AI_MODEL_DIRECTOR=Qwen/Qwen3-Coder-480B-A35B-Instruct
# AI_MODEL_CODEGEN=Qwen/Qwen3-Coder-480B-A35B-Instruct
# AI_MODEL_FBI=Qwen/Qwen3-Coder-480B-A35B-Instruct

# Groq Configuration (for core/main.js)
GROQ_API_KEY=your_groq_key

# Zoom RTMS Configuration (for core/main.js)
ZOOM_CLIENT_ID=your_zoom_client_id
ZOOM_CLIENT_SECRET=your_zoom_client_secret
ZOOM_SECRET_TOKEN=your_zoom_secret_token
WEBHOOK_PATH=/webhook

# Salesforce MCP Configuration (for core/main.js)
SALESFORCE_MCP_URL=your_salesforce_mcp_url
PARALLEL_API_KEY=your_parallel_api_key

AI Model Configuration

ERA supports per-component model and provider configuration. You can mix different providers (Wandb, OpenAI, Groq) for optimal speed/cost/quality.

⚠️ Important: WANDB_API_KEY is always required for Weave tracing, even if you use other providers for inference.

Components: director, codegen, fbi, sluggen, promoter

Configuration Priority:

Models: AI_MODEL_<COMPONENT> → AI_MODEL → Qwen/Qwen3-Coder-480B-A35B-Instruct
URLs: INFERENCE_URL_<COMPONENT> → INFERENCE_URL → https://api.inference.wandb.ai/v1/chat/completions
Keys: INFERENCE_API_KEY_<COMPONENT> → INFERENCE_API_KEY → WANDB_API_KEY

Example 1: Default (Wandb with Qwen3)

WANDB_API_KEY=your_wandb_api_key_here
AI_MODEL=Qwen/Qwen3-Coder-480B-A35B-Instruct

Example 2: Use OpenAI for Everything

WANDB_API_KEY=your_wandb_key_here  # Still needed for Weave tracing!
INFERENCE_URL=https://api.openai.com/v1/chat/completions
INFERENCE_API_KEY=sk-your_openai_key_here
AI_MODEL=gpt-4o

Example 3: Mix & Match (Groq for Speed + OpenAI for Quality)

# Fast operations (Director, FBI, Slug) → Groq
INFERENCE_URL_DIRECTOR=https://api.groq.com/openai/v1/chat/completions
INFERENCE_API_KEY_DIRECTOR=gsk_your_groq_key_here
AI_MODEL_DIRECTOR=llama-3.3-70b-versatile

INFERENCE_URL_FBI=https://api.groq.com/openai/v1/chat/completions
INFERENCE_API_KEY_FBI=gsk_your_groq_key_here
AI_MODEL_FBI=llama-3.3-70b-versatile

INFERENCE_URL_SLUGGEN=https://api.groq.com/openai/v1/chat/completions
INFERENCE_API_KEY_SLUGGEN=gsk_your_groq_key_here
AI_MODEL_SLUGGEN=llama-3.1-8b-instant

# Quality code generation → OpenAI
INFERENCE_URL_CODEGEN=https://api.openai.com/v1/chat/completions
INFERENCE_API_KEY_CODEGEN=sk-your_openai_key_here
AI_MODEL_CODEGEN=gpt-4o

# Fallback for promoter → Wandb
WANDB_API_KEY=your_wandb_key_here

Deno Configuration

The deno.json file contains:

Tasks - Predefined commands for common operations
Imports - Dependency mapping (npm and jsr packages)
Compiler Options - TypeScript configuration
Deploy Settings - Deno Deploy configuration

📚 Available Tasks

Task	Description
`deno task cli`	Run CLI in interactive mode
`deno task cli:create`	Create an agent (with args)
`deno task dev`	Run web server with watch mode
`deno task serve`	Run production web server
`deno task prod`	Deploy to Deno Deploy
`deno task test:daytona`	Test Daytona integration
`deno task test:wandb`	Test Wandb integration
`deno task test:weave`	Test Weave integration
`deno task git`	Quick git commit and push

🧪 Tech Stack

CLI

Deno Standard Library - File operations, path handling
Custom ANSI Colors - Terminal styling (no external deps)
Custom Banner - ASCII art generation

Web Server

Hono - Fast web framework for Deno
Alpine.js - Lightweight reactive framework
Tailwind CSS - Utility-first CSS framework

Backend

Daytona SDK - Code sandbox execution
Wandb - AI experiment tracking and inference
Weave - AI operation tracing
OpenAI - LLM inference (via Wandb and Groq)

🧠 How It Works: The FBI Director System

ERA uses an intelligent "FBI Director" system that orchestrates code generation:

Workflow

Director Reviews - Analyzes your prompt and improves it with:
- Specific technical requirements
- Expected input/output behavior
- Error handling considerations
- Best practices
Code Generation - Creates code using improved prompt via Wandb AI
Sandbox Execution - Tests code in Daytona sandbox environment
Director Verdict - Analyzes results and decides:
- ✅ Success - code works, stop here
- 🔄 Retry - has errors, refine and try again
- 🛑 Stop - no progress, accept current version
Iteration - If verdict is retry, Director refines the prompt based on errors and repeats

Interactive Refinement

After successful generation, you can provide natural language feedback:

Would you like to refine/improve this agent? (y/n): y

What would you like to improve?: Add input validation and better error messages

The Director will:

Understand your feedback
Build context from previous successful code
Generate an improved version
Test it in the sandbox
Ask if you want to refine further

Iteration History

Every attempt is preserved:

agent.json contains ALL attempts with prompts, code, and results
iterations/ folder has timestamped snapshots of each version
index.ts always has the latest/best version

This means:

✅ Never lose work
✅ Review what changed between iterations
✅ Rollback if needed
✅ Learn from the progression

🎯 Examples

Example 1: Create and Refine an Agent

$ deno task cli

🚀 Welcome to ERA CLI!

What do you want the agent to do?: Create a function that reverses a string

🤖 Generating agent name...
Agent name: (default: string-reverser) [press Enter]

🤖 Creating agent: string-reverser
💬 Prompt: Create a function that reverses a string

⏳ Calling ERA orchestrator...
✅ Success! Code generated and validated

Would you like to refine/improve this agent? (y/n): y

What would you like to improve?: Add unicode emoji support and handle edge cases

🔄 Refining agent with your feedback...
✅ Refinement complete!

Would you like to refine/improve this agent? (y/n): n

✅ Agent finalized!

Result:

agents/string-reverser/index.ts - Your refined agent
agents/string-reverser/agent.json - Full history of both attempts
agents/string-reverser/iterations/ - Snapshots of each version

Example 2: Review Iteration History

# Run the latest version
deno run agents/string-reverser/index.ts

# Compare iterations
cat agents/string-reverser/iterations/iteration-1-*.ts
cat agents/string-reverser/iterations/iteration-2-*.ts

# View complete metadata
cat agents/string-reverser/agent.json

Example 3: Agent Continuation (Returning Later)

# Session 1: Create initial agent
$ deno task cli
What to do?: Create a calculator
Agent name: calculator
✅ Created!

# ... Later (hours, days, weeks) ...

# Session 2: Continue development
$ deno task cli
What to do?: Add division and modulo operations
Agent name: calculator  # Same name!

📂 Found existing agent: calculator
   Previous attempts: 1
   Continuing with new iteration...

✅ Updated! Now has division and modulo

# View session history
$ cat agents/calculator/agent.json | jq '.sessions'
[
  {
    "timestamp": "2025-10-12T02:00:00.000Z",
    "prompt": "Create a calculator",
    "attemptCount": 1
  },
  {
    "timestamp": "2025-10-12T03:30:00.000Z",
    "prompt": "Add division and modulo operations",
    "attemptCount": 1
  }
]

Example 4: Start the Web Server

deno task dev
# Visit http://localhost:8000

🔐 Permissions

Deno is secure by default. This project requires:

--allow-net - Network access (for web server and API calls)
--allow-read - File system reads (for HTML, config files)
--allow-write - File system writes (for creating agents)
--allow-env - Environment variable access (for API keys)
--allow-import - Dynamic imports (for modules)

These are pre-configured in deno.json tasks.

🚢 Deployment

Deno Deploy

# Deploy to production
deno task prod

The deploy configuration is in deno.json under the deploy key.

Other Platforms

The project also supports:

Val.town - Set valtown env var
Smallweb - Compatible out of the box
Self-hosted - Use deno task serve

📝 Development

Project Philosophy

Deno-Native - No Node.js, no npm, pure Deno
Zero Config - Works out of the box
Type Safe - Full TypeScript support
Minimal Dependencies - Leverage Deno standard library

Adding New Features

Backend Modules - Add to core/backend/
Frontend Pages - Add to core/frontend/
CLI Commands - Extend cli.ts

Code Style

Use Deno fmt: deno fmt
Use Deno lint: deno lint

📚 Documentation

docs/WORKFLOWS.md - Complete workflows: experimental agents, stable utilities, AI-powered promotion
docs/ARCHITECTURE.md - System design: core/, utils/, agents/ separation and philosophy
docs/UTILITY-PROMOTION.md - Deep dive into the promotion system and examples.ts generation
docs/NPM-PACKAGES-GUIDE.md - How generated agents can use npm packages in Daytona
docs/README-UTILS-INJECTION.md - Utils injection system

Quick Reference

# Workflows
deno task test:create-util    # Guide for creating utilities
deno task test:promotion       # Guide for promoting agents

# System
deno task test:registry        # See all available utilities
deno task test:e2e-agent       # Test self-improving loop

🤝 Contributing

Fork the repository
Create a feature branch
Make your changes
Run deno fmt and deno lint
Submit a pull request

📄 License

ISC

🙏 Acknowledgments

Built with Deno
Powered by Hono
Styled with Tailwind CSS
UI with Alpine.js

📞 Support

Open an issue on GitHub
Check the Deno documentation
Visit Hono documentation

Made with ❤️ using Deno

Name		Name	Last commit message	Last commit date
Latest commit History 88 Commits
core		core
docs		docs
tests		tests
utils		utils
web		web
.dockerignore		.dockerignore
.eslintrc.json		.eslintrc.json
.gitignore		.gitignore
.npmrc		.npmrc
.prettierrc		.prettierrc
Dockerfile.web		Dockerfile.web
README.md		README.md
cli.ts		cli.ts
deno.json		deno.json
env.example		env.example
fly.toml		fly.toml
history.ts		history.ts
main.ts		main.ts

Folders and files

Latest commit

History

Repository files navigation

ERA 🚀

Try us now!

🌟 Self-Improving Meta-Programming System

Features

🚀 Quick Start

Prerequisites

Installation

📦 Project Structure

Agent Creation Process (How it works)

🔧 Usage

CLI Tool - Create Agents

Interactive Mode (Recommended) 🆕

Self-Improving Example 🌟

Quick Start

Command Line Mode

Generated Files Structure

Running Your Agents

Web Server - Hono Application

Development Mode (with hot reload)

Production Server

Deploy to Deno Deploy

Backend Modules

Test Daytona Integration

Test Wandb Integration

Test Weave Integration

⚙️ Configuration

Environment Variables

AI Model Configuration

Example 1: Default (Wandb with Qwen3)

Example 2: Use OpenAI for Everything

Example 3: Mix & Match (Groq for Speed + OpenAI for Quality)

Deno Configuration

📚 Available Tasks

🧪 Tech Stack

CLI

Web Server

Backend

🧠 How It Works: The FBI Director System

Workflow

Interactive Refinement

Iteration History

🎯 Examples

Example 1: Create and Refine an Agent

Example 2: Review Iteration History

Example 3: Agent Continuation (Returning Later)

Example 4: Start the Web Server

🔐 Permissions

🚢 Deployment

Deno Deploy

Other Platforms

📝 Development

Project Philosophy

Adding New Features

Code Style

📚 Documentation

Quick Reference

🤝 Contributing

📄 License

🙏 Acknowledgments

📞 Support

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages