Your AI agent should see more than you can. Not just do more.
Waves™ 2.0 is the first Product Consciousness Framework. It converts AI agents from fast executors into strategic advisors with graduated autonomy — agents that know your business, enforce your rules mechanically, classify every decision by impact level, and alert you to what you can't see because you're focused on the immediate task. From idea validation with Monte Carlo simulations to production code with real-time enforcement.
AI agents compress months of work into days. Fixed 2-week sprints create artificial boundaries around work that flows at a completely different speed. Ceremonies become overhead, not rhythm.
Your AI agent forgets everything between sessions. You re-explain your architecture, your conventions, your decisions. It invents its own patterns. You fix it. Repeat. Every single day.
You have an idea but no structured way to validate it, define the product, plan the roadmap, and track implementation — all with the same AI agent keeping full context.
Your AI agent treats renaming a variable the same as changing your business model. It doesn't know when to proceed and when to stop. It can't distinguish a formatting fix from a decision that affects your product's revenue stream. Without classification, there's no trust — and without trust, you can't truly delegate.
Waves 1.x gave structure to chaos. Waves 2.0 gives consciousness to structure.
Rules That Can't Be Ignored
CLAUDE.md instructions degrade in long sessions — it's a known LLM limitation. Waves 2.0 uses executable hooks (real code, not suggestions) that block actions when artifacts are missing. The agent can't skip your blueprint, can't ignore your roadmap, can't forget your rules.
Graduated Autonomy
Every decision the agent makes is classified by impact: Level 1-2 (technical) → proceed autonomously. Level 3 (scope change) → stop and inform. Level 4 (business impact) → stop and project scenarios. Level 5 (discovery) → stop, document, and advise. The agent knows when to act and when to ask.
The Agent Sees What You Can't
Three automatic triggers: when a primary objective completes, the agent reflects holistically before continuing. When the blueprint changes, it projects cascading impacts on all active work. When a roadmap phase completes, it produces a full strategic audit with recommendations.
Waves answers: it depends on the decision — and the agent knows.
Validated with 18 benchmark scenarios across all 5 levels: 100% correct actions, conservative bias when uncertain (classifies UP, never down). The system is ready for real delegation.
Waves replaces fixed sprints with organic delivery cycles. Each wave carries a product increment from validation to production — lasting as long as it needs. No timeboxes. No artificial ceremonies.
feasibility → foundation → blueprint → roadmap_w0 → logbooks → roadmap_w1 → logbooks
Start at any level. If you already have a product, jump straight to a business wave. The full pipeline is for building from scratch.
Each wave carries a product increment from validation to production. A wave lasts as long as it needs — sometimes 3 days, sometimes 3 weeks. No artificial ceremonies, no arbitrary time boxes.
Most frameworks start at code. Waves starts at the business case. Monte Carlo simulations, market analysis, and Bayesian projections — so you know your idea can work before you write a line.
Same structured files work across Claude Code, Claude Desktop, Codex, and Gemini CLI. Switch agents, switch tools — the context follows. No vendor lock-in.
Academic research, creative projects, business planning — Waves adapts. Different schemas for different project types, same structured workflow.
Waves retains only what matters. No bloated context, no verbose summaries. Structured files keep objectives precise and context lean — so your agents stay fast and accurate across sessions.
In production since December 2024. Evolved from 'ai-behavior' (structure) to Waves 1.x (traceability) to Waves 2.0 (consciousness). v2.0 validated with 29 automated tests: 11 hook tests + 18 classification benchmarks. 100% pass rate.
Traditional frameworks depend on human discipline. Waves depends on code. Hooks execute obligatorily — the agent can't ignore the blueprint, skip documentation, or take business decisions without your input. This isn't a best practice guide. It's executable governance.
The agent doesn't just execute your task — it sees the entire board. When you're focused on implementing a feature, the agent sees how it affects the blueprint, what risks it creates for upcoming phases, and what discoveries have independent market value. You focus on the piece. The agent watches the game.
These aren't projections. This is what we've measured after 15+ months of daily production use.
No more re-explaining architecture, conventions, or previous decisions. The logbook has it all.
Enforced coding rules eliminate style inconsistencies. Code passes review on the first try.
Same structured files work across Claude Code, Claude Desktop, Codex, and Gemini CLI. Switch tools, keep context.
Feasibility analysis catches bad ideas before you invest. Foundation and blueprint ensure you build the right thing, not just build it right.
Decision classification validated across all 5 levels with real product scenarios. The agent never proceeded when it should have stopped.
brew tap exovian-developments/waves
brew install waves
# Initialize in your project
cd your-project
waves init
# Then in Claude Code:
/waves:project-init
Download waves.plugin from GitHub Releases and double-click to install in Claude Desktop. The plugin auto-loads your project context on session start.
waves.plugin from GitHub Releases
/waves:project-init to configure your project
git clone https://github.com/exovian-developments/waves.git
# Copy commands to your project
mkdir -p your-project/.claude/commands
cp -r waves/.claude/commands/* your-project/.claude/commands/
# Copy schemas to your project
mkdir -p your-project/ai_files/schemas
cp waves/schemas/*.json your-project/ai_files/schemas/
Run these two commands to get your project fully configured:
/waves:project-init
Sets up your preferences and project context
/waves:manifest-create
Analyzes your codebase and creates the project manifest
From this point on, every AI agent that reads your project will understand its architecture, conventions, and history.
Hooks-based features (enforcement, metacognition, classification) are currently exclusive to Claude Code. Schema-based features (artifacts, persistence, commands) work on all platforms.
Waves 2.0 — Product Consciousness Framework. In production since 2024. 29/29 tests passed. Install in 60 seconds. Free forever. Open source. Apache 2.0.