ROADMAP

Caxton Roadmap

This document outlines the development phases for Caxton, with concrete milestones and explicit non-goals. It is aligned with the comprehensive user story backlog in PLANNING.md.

Vision

Build the simplest possible server for multi-agent systems, then let the community build amazing agents and workflows.

Development Phases

Phase 1: Minimal Core (V1.0) - Q1 2025

Goal: Ship a production-ready orchestration server with baseline performance.

User Stories: P0 (Stories 001-004) + P1 (Stories 005-010)

Critical foundation components
Essential features for MVP

Deliverables:

WASM agent runtime with isolation guarantees (Story 001)
Async message router with FIPA ACL support (Stories 002, 005)
Agent lifecycle management (Story 003)
Local SQLite state storage (Story 004)
External agent routing API (gRPC + REST gateway) (Stories 006, 007)
CLI tool for agent deployment and management (Story 008)
OpenTelemetry integration (traces, metrics, logs) (Story 009)
Basic MCP tool integration (Story 010)
Health checks and readiness probes (Story 017)

Success Metrics:

Time to first agent: < 10 minutes
External API latency: < 1ms overhead for local calls
Message throughput: 100K/sec minimum
Memory per agent: < 10MB
Zero message loss under normal operation
API error responses include actionable debugging information

Explicitly NOT in V1:

Distributed clustering
Complex routing patterns
Built-in agent templates
Graphical debugging tools
Performance optimizations beyond baseline

Phase 2: Patterns & Performance (V2.0) - Q3 2025

Goal: Double performance and provide pattern libraries (not in core).

User Stories: P2 (Stories 011-017) + Selected P3 (Stories 018-025)

Standard production features
Enhanced security and operations

Deliverables:

Success Metrics:

2x performance improvement (200K messages/sec)
50% reduction in memory usage
External API supports streaming and batch patterns
Pattern library covers 80% of use cases
Agent development possible in 4+ languages
Production-ready security and auth integrations
< 30 second cluster convergence
Zero-downtime deployments

Still NOT in scope:

Infrastructure-level consensus (Raft, Paxos, PBFT)
Built-in workflow engine
Agent hierarchies
Automatic scaling

Phase 3: Scale & Ecosystem (V3.0) - Q1 2026

Goal: Enable planet-scale agent systems through composability.

User Stories: Remaining P3 + P4 (Stories 026-040)

Advanced scaling features
Enterprise capabilities
Ecosystem enablement

Deliverables:

Success Metrics:

10x scale increase from V1
99.99% availability in production
100+ community-contributed patterns
Major cloud providers offer Caxton
Complete multi-tenant isolation
Plugin ecosystem thriving
Enterprise compliance certified

Permanent Non-Goals

We will NEVER add these to Caxton core:

Complex Orchestration

❌ Workflow definition languages
❌ BPMN/BPEL support
❌ Visual flow designers
✅ Users can implement these as agents

Agent Management

❌ Built-in permission systems
❌ Agent lifecycle management
❌ Resource quotas and limits
✅ Cloud platforms can add these

Message Transformation

❌ Message routing rules engine
❌ Content-based routing
❌ Protocol translation
✅ Agents can implement these

Distributed Systems Magic

❌ Consensus protocols (Raft/Paxos)
❌ Distributed transactions
❌ Exactly-once guarantees
✅ Be honest about distributed reality

How to Contribute

Phase 1 Priorities

Performance optimizations - Help us hit throughput targets
WASM security - Ensure true isolation
Documentation - Examples and tutorials
Testing - Chaos testing, benchmarks

Phase 2 Opportunities

Pattern development - Share your agent patterns
Language bindings - Make agents easy to build
Debugging tools - Trace visualization and analysis
Performance - Profile and optimize

Phase 3 Challenges

Distributed systems - Clustering and partitioning
Ecosystem - Integrations and tools
Production hardening - Real-world testing
Standards - Work with FIPA and MCP communities

Principles

Simplicity wins - Every feature must pay for its complexity
Performance matters - Fast by default, optimize later
Observability first - If you can’t debug it, it’s broken
Composition over features - Simple capabilities that combine well
Honest about tradeoffs - No distributed systems magic

Implementation Planning

For detailed implementation planning, see PLANNING.md which contains:

40 comprehensive user stories with full acceptance criteria
Complete Definition of Done for each story
Priority-based backlog (P0-P4)
Coverage matrix mapping all ADRs to stories
Success metrics alignment

Each story in PLANNING.md represents a complete vertical slice that can be independently developed, tested, and released.

Questions?

Join our discussions:

GitHub Discussions for design decisions
Discord for real-time chat
Monthly community calls

Remember: The best server is one that gets out of your way. Let’s build that together.