Core Philosophy

We're extending Cloudflare's vision for AI Agents with a focus on:

• 🤝 Interoperability First: AI Agents as teammates and organizations that can generate revenue and perform advanced operations
• 💰 Cost-Effective Hosting: Shared hosting options with edge-optimized performance
• 🔒 Security by Design: Secure sensitive assets like trading agents and treasuries
• 📈 Self-Improvement: Agents that evolve based on collective usage patterns
• 🔓 No Vendor Lock-in: Full self-hosting and personal account options

Our goal is to empower laymen users to effortlessly self host AI Agents, MCP tools and the data/context that supercharges the experience while rewarding contributors & developers who have contributed to making that possible.

Architecture

The framework is built around several core implemented components:

Sessions

Multi-session support with robust authentication patterns enables:

• Persistent Context: Maintain conversation state across interactions
• User Management: Handle authentication and authorization seamlessly
• Session Isolation: Secure separation between different user sessions

Agents made Easy

Deep integration with modern AI tooling provides:

• Model Agnostic: Work with any LLM provider through standardized interfaces
• Tool Calling: Native support for function calling and tool execution
• Streaming Support: Real-time response streaming for better UX
• Inclusive SDK Support: Support AI SDK, Agents SDK, ElizaOS, and more
• Model Agnostic: Make it dead simple to bring your own keys to any AI model you or your users are keen to try

MCP Tools for Everyone

Ensure MCP Tools are dead simple to use and can run anywhere:

• Cost Effective: MCP Tools run locally when necessary, in the cloud only when used, and reward their builders.
• Play Safely: Ensure tools run in secure boxes on your machines and in their own private clouds that no one has access to.
• Clear Direction: Provide type-safe examples that showcase our opinion on how AI Agents and MCP tools speak to each other and are used by developers.

Services

Ensure Agents can connect, interact, and consume the traditional internet:

• Data Injestion: Enable the ability to injest realtime data from webhooks and 3rd party services
• Admin Mode Online: Create capabilities for super users to manage, tweek and configure the brain of Agents with ease
• Sharing is Caring: Ensure services are reusable packages that can work across multiple use cases if needed

Middleware

Empower developers to customize and react to user messages with ease:

• Tool Injection: Easily configure dynamic tools and inject them when necessary
• Enrich Context: Add or remove context for the AI model to focus on achieving its goal
• Custom Logic: Add or override business logic without disturbing the natural flow of ai models.

Key Features

✅ Ready for Production

• Core MCP Framework: Full Model Context Protocol implementation
• Multi-Session Support: Robust session management with authentication patterns
• WebSocket & HTTP Support: Official MCP WebSocket and HTTP streaming support
• Agent Framework: Complete AI SDK integration for building intelligent agents
• MCP Plugin System: Seamless MCP plugins through mcp.json configuration

🚧 In Active Development

• Framework Integrations: LangChain and additional Agent SDK examples
• Fullstack Examples: Cloudflare Pages with Server-Sent Events
• Advanced Auth: OAuth and JWT implementations

Integration Ecosystem

The framework integrates with modern development tools:

• Getting Started: Build your first agent with step-by-step guidance
• AI SDK: Complete integration with Vercel's AI SDK for model interactions
• ElizaOS: Compatibility layer for ElizaOS agent patterns
• Agent SDK: Native agent development toolkit

Getting Started

The framework is designed for progressive adoption - start simple and scale up:

1. Build Your First Agent: Create a Simple Prompt Agent with session management
2. Add Tools: Integrate external APIs and services through MCP
3. Multi-Agent: Coordinate multiple specialized agents
4. Production: Deploy with full monitoring and scaling

Development Status

This project is in pre-alpha and actively evolving. The framework follows semantic versioning with automated releases:

• Pull Request Testing: Automatic testing and semantic-release dry runs
• Automated Publishing: Changed packages automatically published to npm
• Conventional Commits: Version determination through commit standards

Community & Support

Join our growing community of AI agent developers:

• Discord: Real-time support and discussions
• GitHub: Open source contributions and issue tracking

Built with ❤️ by the Xava DAO Community