██████╗ ██████╗ ███╗ ██╗██╗ ██╗ ██████╗ ██╗ ██╗███████╗
██╔════╝██╔═══██╗████╗ ██║██║ ██║██╔═══██╗██║ ██╔╝██╔════╝
██║ ██║ ██║██╔██╗ ██║██║ ██║██║ ██║█████╔╝ █████╗
██║ ██║ ██║██║╚██╗██║╚██╗ ██╔╝██║ ██║██╔═██╗ ██╔══╝
╚██████╗╚██████╔╝██║ ╚████║ ╚████╔╝ ╚██████╔╝██║ ██╗███████╗
╚═════╝ ╚═════╝ ╚═╝ ╚═══╝ ╚═══╝ ╚═════╝ ╚═╝ ╚═╝╚══════╝
Agent teams for complex systems
Convoke extends AI agents with two types of installable modules: Teams bring new agents for a domain, Skills add new capabilities to existing agents. Both are top-level modules — install them independently or combine them.
Vortex is the first team (product discovery, 7 agents). Enhance is the first skill (RICE backlog management for the PM agent). More of each are coming.
7 agents guide you from insight to evidence and back again — a continuous discovery loop, not a one-shot checklist
Most teams skip validation and build on assumptions.
Vortex guides you through seven discovery streams — from understanding your users to interpreting production signals — so you can make evidence-based decisions before, during, and after you build. Each stream builds on the previous one's findings, and when gaps appear, the system routes you back to fill them.
7 Streams · 7 Agents
┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ Isla │──▶│ Mila │──▶│ Liam │──▶│ Wade │
│ Empathize │ │ Synthesize │ │ Hypothesize │ │ Externalize │
└─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘
▲ │
│ │
│ ▼
┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ Emma │◀──│ Max │◀──│ Noah │◀───────────┘
│Contextualize│ │ Systematize │ │ Sensitize │
└─────────────┘ └─────────────┘ └─────────────┘
│ │ │
└─────────────────┴─────────────────┘
▶ Start at Emma · back to any stream
Suggested flow. Each workflow ends with a Compass routing to whichever stream needs attention — you can start or return to any agent.
22 Vortex Workflows
- Assumption Mapping
- Behavior Analysis
- Contextualize Scope
- Empathy Map
- Experiment Design
- Hypothesis Engineering
- Lean Experiment
- Lean Persona
- Learning Card
- MVP
- Pattern Mapping
- Pivot Patch Persevere
- Pivot Resynthesis
- Product Vision
- Production Monitoring
- Proof of Concept
- Proof of Value
- Research Convergence
- Signal Interpretation
- User Discovery
- User Interview
- Vortex Navigation
| Agent | Stream | What they do |
|---|---|---|
| Emma 🎯 | Contextualize | Frame the right problem — personas, product vision, scope |
| Isla 🔍 | Empathize | Understand users — empathy maps, interviews, discovery research |
| Mila 🔬 | Synthesize | Converge research into clear problem definitions |
| Liam 💡 | Hypothesize | Turn problems into testable hypotheses and experiments |
| Wade 🧪 | Externalize | Test assumptions with MVPs, experiments, and prototypes |
| Noah 📡 | Sensitize | Interpret production signals, user behavior, and engagement patterns |
| Max 🧭 | Systematize | Capture learnings and decide: pivot, patch, or persevere |
Here's a sample of real output from a busy parents meal planning project — each excerpt is from the full 7-agent journey example.
Emma frames the right problem. Here's the Job-to-be-Done she produced:
Job-to-be-Done: Eliminate the daily 5:30 PM dinner decision so I can feed my family well without the mental load of planning, shopping, and deciding under time pressure.
Riskiest Assumptions:
- Decision fatigue — not cooking skill or ingredient access — is the primary barrier to weeknight dinner success
- Parents would trust and act on an externally-provided dinner suggestion rather than needing to choose themselves
- "Good enough" nutrition is an acceptable standard — parents aren't seeking perfection, they're seeking relief from guilt
Liam turns problems into testable ideas. Here's one of three hypotheses he produced:
Hypothesis 1: The Pre-Commute Decision Eliminator
We believe that busy parents will act on a single dinner suggestion delivered at 4:00 PM within 3 minutes because the decision burden — not cooking — is their primary barrier, and an earlier intervention catches them before the anxiety spiral begins.
Riskiest Assumption: Parents will trust and act on an automated suggestion without second-guessing. Research shows they want "someone to tell me what to make" — but "someone" may need to be a trusted person, not an algorithm.
Max captures what you learned and decides what to do next:
Recommendation: PATCH (iterate on timing, don't pivot direction)
The core hypothesis is validated. The product direction (decision elimination via single suggestion) is correct. The timing mechanism needs refinement — shift from fixed 4:00 PM delivery to adaptive delivery based on each user's observed engagement pattern.
Three Actions:
- Implement adaptive timing — shift the push notification to match each user's observed engagement window. Engineering effort: 1-2 sprints.
- Route to Isla for timing investigation — qualitative research on why users engage at 3:15 PM. Is it anxiety relief, logistical planning, or habit?
- Test willingness to pay immediately — the mechanism works, but we have no commercial validation. Run a landing page test with pricing before further product investment.
Want to see the complete walkthrough of all 7 agents applied to the example above?
See the full 7-agent journey example →
- Node.js 18+ or Bun
- Git
- Claude Code or Claude.ai
npm install convoke-agents && npx -p convoke-agents convoke-install-vortexAll 7 agents with 22 workflows are installed and ready to use. Something not working? Run npx -p convoke-agents convoke-doctor or check the FAQ.
Open _bmad/bme/_vortex/config.yaml and replace {user} with your name. Agents use this to personalize their interactions.
Claude Code (skills)
/bmad-agent-bme-contextualization-expert # Emma 🎯
/bmad-agent-bme-discovery-empathy-expert # Isla 🔍
/bmad-agent-bme-research-convergence-specialist # Mila 🔬
/bmad-agent-bme-hypothesis-engineer # Liam 💡
/bmad-agent-bme-lean-experiments-specialist # Wade 🧪
/bmad-agent-bme-production-intelligence-specialist # Noah 📡
/bmad-agent-bme-learning-decision-expert # Max 🧭
Claude Code (terminal) / Other AI assistants
cat _bmad/bme/_vortex/agents/contextualization-expert.md # Emma 🎯
cat _bmad/bme/_vortex/agents/discovery-empathy-expert.md # Isla 🔍
cat _bmad/bme/_vortex/agents/research-convergence-specialist.md # Mila 🔬
cat _bmad/bme/_vortex/agents/hypothesis-engineer.md # Liam 💡
cat _bmad/bme/_vortex/agents/lean-experiments-specialist.md # Wade 🧪
cat _bmad/bme/_vortex/agents/production-intelligence-specialist.md # Noah 📡
cat _bmad/bme/_vortex/agents/learning-decision-expert.md # Max 🧭Claude.ai
Open any agent file from _bmad/bme/_vortex/agents/ and paste its contents into your conversation.
How activation works: Each agent is a markdown file containing a full persona, menu system, and workflow instructions. When Claude reads the file, it adopts that agent's expertise and presents you with an interactive menu. Pick a workflow from the menu and follow the guided steps.
- Personalize — If you haven't already, edit
_bmad/bme/_vortex/config.yamland replace{user}with your name - Activate Emma —
cat _bmad/bme/_vortex/agents/contextualization-expert.md - See the menu — Emma presents numbered options. Select Lean Persona from the menu
- Follow the steps — Emma walks you through 6 guided steps, asking questions at each one
- Find your artifact — When done, your lean persona is saved in
_bmad-output/vortex-artifacts/ - Follow the Compass — At the end, Emma's Vortex Compass suggests which agent to use next based on what you learned
Each workflow ends with a Compass routing suggestion. You don't need to follow a linear path — the system guides you to whichever stream needs attention.
your-project/
├── _bmad/bme/
│ ├── _vortex/ # Team: Product Discovery
│ │ ├── agents/ # 7 agent definition files
│ │ ├── workflows/ # 22 workflows
│ │ ├── contracts/ # Artifact contract schemas
│ │ ├── guides/ # User guides (all 7 agents)
│ │ └── config.yaml # Configuration
│ └── _enhance/ # Skill: Agent Capability Upgrades
│ ├── workflows/ # Skill workflows (initiatives-backlog)
│ ├── extensions/ # Agent menu patch descriptors
│ ├── guides/ # Module author guide
│ └── config.yaml # Configuration
└── _bmad-output/
└── vortex-artifacts/ # Generated artifacts
npx -p convoke-agents convoke-version # Check current version
npx -p convoke-agents convoke-update --dry-run # Preview changes
npx -p convoke-agents convoke-update # Apply update (auto-backup)
npx -p convoke-agents convoke-doctor # Diagnose issuesYour data in _bmad-output/ is never touched. Automatic backups are created before every update.
Tip: If
npx convoke-updatereports "Already up to date" but you know a newer version exists, npx may be serving a cached copy. Force the latest with:npx -p convoke-agents@latest convoke-update --yes
See UPDATE-GUIDE.md for migration paths and troubleshooting.
Each agent can be used independently or as part of the full Vortex flow:
- Emma (Contextualize) — Start here when defining a new product or problem space
- Isla (Empathize) — Start here when you need to understand users better
- Mila (Synthesize) — Start here when you have research to converge into a problem definition
- Liam (Hypothesize) — Start here when you have a problem definition and need testable hypotheses
- Wade (Externalize) — Start here when you have hypotheses ready to test
- Noah (Sensitize) — Start here when experiments have graduated to production
- Max (Systematize) — Start here when you have results and need to decide next steps
Max's Vortex Navigation workflow helps identify which stream needs attention based on evidence gaps — you don't have to follow a linear path. Every workflow ends with a Vortex Compass that routes you to the right next agent based on what you learned. Ten handoff contracts (HC1-HC10) ensure structured information flows between agents — see the contracts directory for details.
For detailed workflow descriptions and usage examples, see the Agent Guide and the individual user guides:
- Emma User Guide
- Isla User Guide
- Mila User Guide
- Liam User Guide
- Wade User Guide
- Noah User Guide
- Max User Guide
Add new capabilities to existing agents without modifying them
Skills are the other half of Convoke's extensibility. While Teams bring new agents, Skills give existing agents new workflows — installed via menu patching, not agent modification.
The first Enhance skill adds RICE-scored backlog management to the PM agent. Three modes cover the full lifecycle:
Initiatives Backlog
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ [T] Triage │ │ [R] Review │ │ [C] Create │
│ Ingest new │ │ Rescore │ │ Bootstrap │
│ findings │ │ existing │ │ from scratch│
└──────┬──────┘ └──────┬──────┘ └──────┬──────┘
│ │ │
▼ ▼ ▼
┌──────────────────────────────────────────────────┐
│ initiatives-backlog.md │
│ RICE-scored · Categorized · Change-tracked │
└──────────────────────────────────────────────────┘
| Mode | What it does |
|---|---|
| Triage | Ingest review findings, extract actionable items, propose RICE scores with two-gate validation, append to existing backlog |
| Review | Walk through items one at a time, rescore where priorities have shifted, regenerate prioritized view |
| Create | Gather initiatives interactively, batch-score with RICE, generate a complete backlog from scratch |
The Enhance skill appears in the PM agent's menu after installation:
[IB] Initiatives Backlog (Convoke Enhance)
Select it from the PM agent menu, or activate directly:
Claude Code (skills)
/bmad-enhance-initiatives-backlog
Claude Code (terminal) / Other AI assistants
cat _bmad/bme/_enhance/workflows/initiatives-backlog/workflow.mdThe Enhance Guide documents the complete pattern for creating new skills: directory structure, step file architecture, agent menu patching, config registration, and verification integration. It uses the initiatives-backlog skill as the canonical example throughout.
Max 2-3 skills per agent to prevent menu bloat and maintain agent focus.
Convoke handles discovery and validation. BMAD Core handles implementation.
Convoke Modules BMAD Core
┌──────────────────────────────┐ ┌──────────────────────┐
│ Teams │ │ │
│ Vortex (Product Discovery) │ ──────> │ PM → Architect → Dev │
│ [Future teams] │ │ "Let's build it" │
│ │ <────── │ │
│ Skills │ signals │ │
│ Enhance (Agent Upgrades) │ ──────> │ │
│ [Future skills] │ │ │
└──────────────────────────────┘ └──────────────────────┘
Teams and Skills are peer module types — both installable, both independent. Convoke works standalone or as an extension — no BMAD Method installation required.
| Document | Description |
|---|---|
| Agent Guide | Detailed agent descriptions, workflows, and positioning |
| Testing | Automated test suite, CI pipeline, and agent test results |
| Development Guide | Architecture, building agents, and contributing |
| FAQ | Common questions about the framework |
| CHANGELOG | Complete version history |
| UPDATE-GUIDE | Migration paths and update troubleshooting |
- v1.1.0–v1.5.x — Waves 1-2: Foundation (Emma, Isla, Wade, Max — agent definitions, workflows, update system, CI/CD)
- v1.6.x — Wave 3: Complete 7-stream Vortex (added Mila, Liam, Noah — 7 agents, 22 workflows, handoff contracts, Compass routing)
- v1.7.0 — Wave 4: Quality & onboarding (P0 test suite, docs audit tool, all 22 workflows production-ready, README overhaul, package size fix)
- v2.0.0 — Product renamed to Convoke. CLI commands renamed to
convoke-*. Package:npm install convoke-agents - v2.x — Enhance module: Skills architecture, RICE initiatives-backlog skill (Triage/Review/Create modes), module author pattern guide
- Next — Additional teams, additional skills, multi-agent collaboration, cross-module workflows
We welcome contributions in these areas:
- Agents — New domain-specialized agents, workflow improvements
- Testing — Edge cases, performance testing
- Documentation — Tutorials, translations, video walkthroughs
- Integration — IDE plugins, third-party tools
Have feedback? Found a quality issue, want a missing capability, or have a general comment about an agent? Open an issue and select the Agent/Workflow Feedback template — takes under 30 seconds.
See the Development Guide for architecture details and agent development patterns.
MIT License — see LICENSE
- BMAD Method v6.0.0 — Foundation for agent architecture
- Innovation Vortex — Pattern from the unFIX model by Jurgen Appelo
- Claude (Anthropic) — AI reasoning and agent development
Every agent below contributed to the design, implementation, testing, or documentation of this release.
Core
- BMad Master 🧙 — Orchestration and agent coordination
BMM — BMAD Method Module
- Mary 📊 Analyst — Requirements analysis and product briefs
- Winston 🏗️ Architect — Architecture decisions and technical design
- Amelia 💻 Dev — Implementation and code delivery
- John 📋 PM — Product management and PRD creation
- Barry 🚀 Quick Flow — Rapid solo development
- Quinn 🧪 QA — Quality assurance and test validation
- Bob 🏃 SM — Sprint management and retrospectives
- Paige 📚 Tech Writer — Documentation and editorial review
- Sally 🎨 UX Designer — User experience design
BME — Vortex Pattern Agents
- Emma 🎯 Contextualize — Problem framing and product vision
- Isla 🔍 Empathize — User research and empathy mapping
- Mila 🔬 Synthesize — Research convergence and problem definition
- Liam 💡 Hypothesize — Hypothesis engineering and assumption mapping
- Wade 🧪 Externalize — Lean experiments and MVP testing
- Noah 📡 Sensitize — Production intelligence and signal interpretation
- Max 🧭 Systematize — Learning capture and pivot/persevere decisions
CIS — Creative & Innovation Strategies
- Carson 🧠 Brainstorming Coach — Creative ideation facilitation
- Dr. Quinn 🔬 Problem Solver — Systematic problem-solving
- Maya 🎨 Design Thinking Coach — Human-centered design
- Victor ⚡ Innovation Strategist — Business model innovation
- Caravaggio 🎨 Presentation Master — Visual communication
- Sophia 📖 Storyteller — Narrative and storytelling
BMB — BMAD Builder Module
- Bond 🤖 Agent Builder — Agent creation and configuration
- Morgan 🏗️ Module Builder — Module architecture
- Wendy 🔄 Workflow Builder — Workflow design and validation
TEA — Test Engineering & Architecture
- Murat 🧪 Test Architect — Test framework, CI pipeline, and quality gates
Innovation Vortex Pattern — Discover what has to be discovered, deliver what has to be delivered
Get Started | Agents | Docs | Roadmap