A production-ready starter template for building and deploying AI agents using Mastra on the Nosana decentralized compute network.

This template provides everything you need to build intelligent AI agents with a modern web interface and deploy them on decentralized infrastructure. Built with Mastra, CopilotKit, and Next.js.

• Mastra framework - AI agent orchestration and workflow management
• Tool calling system - Connect your agent to external services and APIs
• MCP (Model Context Protocol) support - Enhanced agent capabilities
• Modern Next.js frontend - Beautiful UI for interacting with your agent
• Docker configuration - Ready for containerized deployment
• Nosana deployment configs - Deploy to decentralized GPU infrastructure

This template can be adapted for various AI agent applications:

• 🤖 Personal Assistant - Schedule management, email drafting, task automation
• 📊 Data Analyst Agent - Fetch financial data, generate insights, create visualizations
• 🌐 Web Researcher - Aggregate information from multiple sources, summarize findings
• 🛠️ DevOps Helper - Monitor services, automate deployments, manage infrastructure
• 🎨 Content Creator - Generate social media posts, blog outlines, marketing copy
• 🔍 Smart Search - Multi-source search with AI-powered result synthesis
• 💬 Customer Support Bot - Answer FAQs, ticket routing, knowledge base queries

• Node.js 18+ and pnpm
• Docker (for deployment)
• Git

# Clone this repository
git clone https://github.com/YOUR-USERNAME/nosana-mastra-template
cd nosana-mastra-template

# Copy environment variables
cp .env.example .env

# Install dependencies
pnpm i

# Start the development servers
pnpm run dev:ui      # Start UI server (port 3000)
pnpm run dev:agent   # Start Mastra agent server (port 4111)

Open http://localhost:3000 to see your agent frontend. Open http://localhost:4111 to access the Mastra Agent Playground.

Choose your preferred LLM provider to power your agent:

Run Ollama locally (requires Ollama installed):

ollama pull qwen3:0.6b
ollama serve

Update your .env:

OLLAMA_API_URL=http://127.0.0.1:11434/api
MODEL_NAME_AT_ENDPOINT=qwen3:0.6b

Add your OpenAI API key to .env and uncomment the OpenAI configuration in src/mastra/agents/index.ts:

OPENAI_API_KEY=your-openai-api-key

Use any OpenAI-compatible endpoint:

OLLAMA_API_URL=https://your-custom-endpoint.com/api
MODEL_NAME_AT_ENDPOINT=your-model-name

Implement your agent logic in the Mastra framework:

1. Define your tools - Create custom functions in src/mastra/tools/
2. Configure your agent - Update agent behavior in src/mastra/agents/
3. Test locally - Validate functionality at http://localhost:3000 and http://localhost:4111

Modify the Next.js frontend to match your agent's functionality:

• Update UI components in src/app/
• Customize the chat interface
• Add custom visualizations or controls

Package your agent for deployment:

# Build your Docker container
docker build -t yourusername/nosana-mastra-agent:latest .

# Test the container locally
docker run -p 3000:3000 yourusername/nosana-mastra-agent:latest

# Push to Docker Hub
docker login
docker push yourusername/nosana-mastra-agent:latest

The provided Dockerfile bundles:

• Your Mastra agent
• Frontend interface
• LLM runtime (all-in-one container)

Deploy your containerized agent to Nosana's decentralized GPU network (see deployment section below).

├── src/
│   ├── app/              # Next.js frontend
│   ├── mastra/           # Mastra agent configuration
│   │   ├── agents/       # Agent definitions
│   │   └── tools/        # Custom tool implementations
│   └── lib/              # Shared utilities
├── nos_job_def/          # Nosana deployment configs
├── Dockerfile            # Container configuration
└── .env.example          # Environment variables template

Deploy your AI agent to Nosana's decentralized GPU network for production use.

1. Open Nosana Dashboard
2. Click Expand to open the job definition editor
3. Edit nos_job_def/nosana_mastra_job_definition.json with your Docker image:

   {
     "image": "yourusername/nosana-mastra-agent:latest"
   }

4. Copy and paste the edited job definition into the dashboard
5. Select your preferred GPU type
6. Click Deploy

Install the Nosana CLI and deploy from your terminal:

# Install Nosana CLI
npm install -g @nosana/cli

# Deploy your agent
nosana job post --file ./nos_job_def/nosana_mastra_job_definition.json --market nvidia-3090 --timeout 30

The job definition file includes:

• Docker image reference
• Resource requirements (GPU, memory, CPU)
• Network exposure settings
• Environment variables

Modify nos_job_def/nosana_mastra_job_definition.json to customize your deployment.

• Decentralized Infrastructure - Run on distributed GPU nodes worldwide
• Cost-Effective - Competitive pricing for GPU compute
• Censorship-Resistant - No single point of control or failure
• Scalable - Easy to scale your agent across multiple nodes
• Transparent - On-chain job execution and verification

• Mastra Documentation - Complete guide to the Mastra framework
• Mastra Agents Overview - Understanding AI agents
• Mastra Tool Calling - Implementing custom tools
• Build an AI Stock Agent Guide - Complete tutorial
• CopilotKit Documentation - Frontend AI integration

• Nosana Documentation - Complete Nosana platform guide
• Nosana CLI - Command-line deployment
• Nosana SDK - JavaScript SDK

• Next.js Documentation - Next.js features and API
• Docker Documentation - Container best practices

Need help or want to connect with other builders?

• Discord - Join Nosana Discord for technical support
• Twitter/X - Follow @nosana_ai for updates
• GitHub - Report issues or contribute to the repos

Contributions are welcome! Feel free to:

• Submit bug reports or feature requests
• Improve documentation
• Share your agent implementations
• Contribute code improvements

This template is open source and available under the MIT License.

Built with Mastra • Deployed on Nosana • Powered by decentralized AI