How Genesis Workflow Studio works

Genesis is a platform for building, running, and governing teams of AI agents. A Workspace is where your team's Projects, AI Teams, Blueprints, and Mission history all live.

Workflow Studio is where you design the *shape* of AI-driven work before it runs — the steps, their order, and what happens if something fails — using a reusable, versioned object called a Blueprint. Mission Runtime is where a Blueprint is actually executed. Studio is the design surface; Runtime is the execution engine.

Workspace
  ├── Dashboard          Overview: activity, status, quick actions
  ├── Projects           Human-facing project & task tracking
  ├── AI Teams           Define who does the work (roles, tools)
  ├── Work Intelligence  Mission planning & requirement clarification
  ├── Safety Authority   Governance policies & approval routing
  ├── AI Providers       AI model provider connections
  ├── Mission Runtime     Every Mission Run, live and historical
  └── Workflow Studio
         ├── Blueprint Library   Design & manage Blueprints
         └── Template Library    Certified, reusable Blueprints

A Blueprint is designed in Workflow Studio. Publishing it creates an immutable, numbered version. A Mission then executes that exact version in Mission Runtime — the Draft can keep changing without affecting a run already in progress.

• Blueprint Library — browse, create, and manage this workspace's Blueprints. Implemented
• Template Library — browse certified and shared Templates, clone one into your own Blueprint. Implemented
• Visual Builder — the canvas editor for a Blueprint's steps. Implemented
• Inspector Panel — edit a selected step's name and its optional Compensation pairing. Partially Implemented (only Name, Compensation, and Sub Workflow's target are editable today — see Visual Builder below)
• Version Compare — see exactly what changed between two Published Versions. Implemented
• Mission Runtime — every run, live or historical, reachable from the main sidebar. Implemented
• Workflow Debugger — step-by-step inspection of one specific, real Mission Run, reached from that run's detail page. Implemented

Create Workspace → Create Blueprint → Design (Visual Builder)
      → Validate → Compare Versions → Publish
      → Mission Runtime (run it) → Workflow Debugger (inspect it)
      → Production Execution

Every stage above is fully usable today.

Steps are added by clicking a type in the Node Catalog (drag-and-drop is Future Feature). Only a step's Name and its Compensation Step pairing are editable for every type; Sub Workflow additionally lets you pick its target Blueprint. Every other type's detailed configuration (a Decision's condition, a Loop's source, etc.) is not yet visually editable.

Draft ──Publish──► Published (v1, v2, v3 …) ──► Deprecated ──► Retired
  ▲                        │
  └──── keep editing ──────┘  (Published versions are immutable)

The Draft is always editable and never versioned directly. Publishing creates an immutable, numbered snapshot. Deprecate/Retire actions are available on any Published Version. Partially Implemented — a full, formal multi-stage Draft → Review → Testing → Approval chain is not fully built; publishing today only requires passing validation.

The Debugger inspects one real, already-started Mission Run — a Timeline of every step, a Prompt Inspector (the exact resolved instruction sent to the agent), a Variable Inspector (input/output data), and an Exception Viewer if a step failed. All Implemented.

Breakpoints Partially Implemented — pause your own view of a live-streaming run, not real execution.

Simulation Mode, Agent Memory Viewer, Tool Response Viewer, and Live Variable Editing are all Future Feature.

Pick any two Published Versions and see exactly what was added, removed, or changed — steps matched by position, not name, so a rename shows as a change, not an add-plus-remove. Implemented

Any step can be paired with another step in the same version as its declared "undo" action. Implemented for authoring, validating, and saving this pairing.

Important: Future Feature — Mission Runtime does not yet automatically invoke a step's compensation when it fails. Treat it as documented intent, not an automatic safety net, until this is built.

Every user has one role per Workspace, from an ordered hierarchy:

• Guest — read-only access.
• Member — everyday work: Blueprints, Missions.
• Admin — workspace configuration.
• Owner — full workspace control.

Starting a Mission creates a Mission Run against one specific Published Blueprint Version. Runs update live as steps execute, can be retried or cancelled, and can pause for Human Approval. All Implemented.

• Compensation Pairing does not execute automatically.
• Most node types can be placed but not configured in detail.
• There is no Simulation Mode — every run is real.
• The Debugger cannot show agent memory or raw tool responses.
• Debugger breakpoints do not pause real execution.
• No drag-and-drop step placement, and no undo/redo in the Builder.

Can I edit a Published Version? No — they're immutable. Edit the Draft and publish a new version instead.

Will my Compensation Step run automatically if something fails? Not yet — this is a documented future capability.

Can I test a Blueprint safely without a real run? Not yet — Simulation Mode is not implemented.