Status: 🚫 NOT PRODUCTION READY - Early Stage Research Project
Kairos is an experimental framework exploring AI-driven collaborative development (also known as "vibe coding"). This project investigates how AI can effectively integrate into team-based software development workflows, not just individual developer-AI interactions.
The goal of Kairos is to:
- Validate AI Integration: Test whether AI-driven development can produce production-grade code through collaborative workflows
- Build a Production Framework: Create a Go-native, observable, and interoperable agent framework that could serve as a foundation for production systems
- Democratize AI Development: Move beyond single-developer-with-AI setups to entire teams leveraging AI as part of their development process
- Establish Best Practices: Define patterns, conventions, and architectural principles for team-based AI development
- A research and experimentation project for AI-assisted development
- A proof-of-concept for team-integrated AI workflows
- An architectural exploration of agent-based systems in Go
- A vibe coding experiment where most development is AI-driven with human oversight
- Production-ready software (APIs are unstable, behavior may change dramatically)
- A finished framework (core components are still being designed and refined)
- Suitable for critical systems (not battle-tested or security-hardened)
- A replacement for traditional development (it's an exploration of new possibilities)
This project operates under a unique development model:
- AI-Driven Development: The majority of code is generated and structured by AI
- Human Oversight: Humans provide direction, validation, and architectural decisions
- Team Integration: Designed to work with entire development teams, not solo developers
- Learning Loop: Each iteration improves both the framework and the AI collaboration process
The ultimate goal is to understand whether this approach can produce enterprise-grade software and to establish patterns that teams can adopt.
kairos/
├── cmd/ # Command-line tools and entry points
├── pkg/ # Core framework packages
├── examples/ # Example usage and integrations
├── docs/ # Technical documentation and ADRs
├── docs-site/ # MkDocs site for documentation
├── scripts/ # Build and utility scripts
├── tools/ # Development tooling
└── AGENTS.md # Guidelines for AI agents working on this project
- Go 1.21+
- Make (optional but recommended)
go build ./...go test ./...go run ./examples/01-hello-agentProduction-Grade Components (Phases 1-3 Complete ✅):
- Error Handling: Typed errors, retry logic, circuit breaker
- Resilience Patterns: Health checks, timeouts, fallbacks
- Observability: 5 OTEL metrics, 6 alerts, 3 dashboards
- 62 tests, 100% pass rate | Zero compiler warnings
See Error Handling Guide for details and examples.
- 👨💻 Developers: Start with Error Handling Guide → Integration Guide for agents → Examples
- 👨💼 Operators: Go to Observability Guide (dashboards, alerts) → Metrics Export Guide (OTLP setup)
- 🏗️ Architects: Read Narrative Guide → ADR 0005
-
Error Handling Guide - Production-grade error handling (All Phases ✅)
- For agents: Integration Guide - How to use error handling in agent loops
- Quick reference: Status | Roadmap | Index
- For executives/architects: Narrative Guide
-
Observability Guide - Dashboards, alerts, monitoring setup
- For operators: Metrics Export Guide - OTLP configuration and where metrics go
- 5 metrics, 6 alert rules, 3 dashboards, PromQL queries included
-
Functional Specification - Complete feature specification (Spanish)
-
Architecture Guide - System design and components
-
API Reference - API documentation
-
CLI Guide - Command-line interface
-
Architecture Decision Records - Design decisions and rationale
-
AGENTS.md - Contributing guidelines for AI-driven development
Kairos implements a reactive agent loop supporting:
- Tool integration (MCP protocol)
- Memory systems for context
- Observable metrics and logging
- Agent discovery and composition
- Agent Loop: ReAct-inspired architecture for agent reasoning
- Tool System: MCP-compatible tool definition and execution
- Memory Management: Persistent and ephemeral context storage
- Observability: Built-in metrics and tracing
See Architecture Documentation for details.
- 🔴 APIs are unstable and may change without warning
- 🔴 Tool ecosystem is minimal
- 🔴 No security hardening (use with caution)
- 🟡 Documentation is incomplete
- 🟡 Performance not optimized
- ✅ Production-grade error handling and observability (see docs and observability guide)
- Stabilize core APIs
- Production-grade error handling and recovery
- Comprehensive tool library
- Security audit and hardening
- Performance benchmarking and optimization
- Integration patterns for team workflows
- Formal validation and testing
This project welcomes contributions and experiments! However, please be aware:
- Architecture First: Before implementing, read AGENTS.md and existing documentation
- Consistency Matters: Follow established patterns in the codebase
- Backward Compatibility: Prefer non-breaking changes when possible
- Documentation: Document architectural decisions in
docs/internal/adr/
See AGENTS.md for detailed contribution guidelines.
The core philosophy behind Kairos:
"Building production software through human-AI collaboration at the team level, where AI augments human judgment rather than replacing it."
Key principles:
- Transparency: All AI-generated code is human-reviewable
- Iterative: Short feedback loops between humans and AI
- Collaborative: Team-based decision making, not solo development
- Learning: Each iteration improves both code and process
- Ambitious: Aim for production-grade output despite experimental nature
[Add your license here]
For questions about the project:
- Check existing documentation
- Review Architecture Decision Records
- See AGENTS.md for AI development guidance
This codebase is actively evolving. Major changes may happen between commits:
- ✅ Learn from it
- ✅ Experiment with it
- ✅ Provide feedback
- ❌ Do NOT use in production systems
- ❌ Do NOT rely on API stability
The goal is eventual production readiness, but we're not there yet.
Made with AI-assisted development | Designed for team collaboration