Caxton Logo

Caxton Documentation

Welcome to the comprehensive documentation for Caxton, the production-ready multi-agent application server.

What is Caxton?

Caxton is a specialized server for coordinating smart software components called “agents” - like Redis for caching or Nginx for web serving, but for managing intelligent automation. You install Caxton, deploy your agents (written in any language that compiles to WebAssembly), and it handles all the complex distributed systems challenges: message routing, fault tolerance, observability, and scaling - all without requiring any external databases or dependencies.

What problems does Caxton solve?

Building intelligent automation is hard. Here are common challenges Caxton addresses:

Coordinating Multiple AI Systems

  • Problem: You have different AI models (LLMs, vision models, specialized algorithms) that need to work together
  • Solution: Caxton routes messages between agents, so your GPT-4 agent can request image analysis from your vision agent

Scaling Smart Applications

  • Problem: Your AI application works great for one user, but breaks under load or when processing multiple requests
  • Solution: Caxton handles resource management, queuing, and scaling so your agents stay responsive

Integrating with Existing Systems

  • Problem: Your AI agents need to read databases, call APIs, send emails, or interact with other services
  • Solution: Caxton’s tool system (MCP bridges) lets agents safely interact with external systems

Making AI Systems Observable

  • Problem: When your AI application misbehaves, you can’t see what went wrong or why
  • Solution: Caxton provides built-in tracing and monitoring so you can debug distributed AI workflows

Real-World Examples

  • Customer Service: Route inquiries between specialist agents (billing, technical, sales)
  • Content Processing: Coordinate agents that analyze, summarize, and moderate user-generated content
  • E-commerce: Have agents handle inventory, recommendations, and fraud detection working together
  • Data Analysis: Pipeline where agents clean data, run analysis, and generate reports
🚀

Quick Start

Get up and running with Caxton in 5 minutes

 🤖

Build Your First Agent

Step-by-step guide to creating and deploying agents

 📖

API Reference

Complete REST and gRPC API documentation

 🏛️

Architecture Decisions

Understand the design choices behind Caxton

Core Concepts

Agents

Think of agents as small, focused programs that handle specific tasks. For example:

  • A billing agent that processes payments and invoices
  • A notification agent that sends emails and texts
  • A data agent that reads from databases and APIs

Each agent runs in its own secure sandbox with configurable memory and CPU limits, like lightweight containers but even safer.

Messages

Agents communicate by sending structured messages to each other, similar to REST API calls but between your own components. For example:

{
  "type": "request",
  "from": "customer-service",
  "to": "billing-agent",
  "content": "Please create invoice for order #12345"
}

Caxton uses a simplified version of industry-standard agent messaging protocols, keeping the useful parts (reliable delivery, request tracking) while discarding academic complexity. See ADR-0012 for our pragmatic approach.

Observability

When your agents are working together, you need to see what’s happening. Caxton provides:

  • Distributed tracing: Follow a customer request across multiple agents
  • Performance metrics: See which agents are slow or failing
  • Structured logging: Debug issues with detailed, searchable logs

All built-in from day one using industry-standard OpenTelemetry.

Tools (MCP Integration)

Agents often need to interact with external systems. Rather than giving agents direct database access (risky), Caxton provides controlled “tools” they can use:

  • Database queries through secure connections
  • API calls with rate limiting and error handling
  • File system access with proper permissions
  • Integration with services like Slack, GitHub, or your internal APIs

This uses the Model Context Protocol (MCP) standard for safe, observable external interactions.

Documentation Structure

Getting Started

Developer Resources

Operations & Production

Architecture & Design

Contributing

Need Help?

💬 Community Support

Join our GitHub Discussions to ask questions and share experiences

🐛 Report Issues

Found a bug? Report it on our Issue Tracker

📺 Examples

Check out example agents and use cases in our examples directory

Edit this page on GitHub →