How Qubitz Works

Qubitz moves you through four stages, from understanding where you stand to running agent systems in production. The last stage returns to the first - this is a cycle you go around once per use case, not a one-way pipeline.

The Four Stages

You do not have to start at the beginning. Choose Your Path covers the three entry points and when each makes sense.

Envision

One assessment run establishes where you stand and produces the use cases to choose from. Readiness and opportunity come out of the same pass, which is why they are one stage rather than two.

Assessing readiness

Click AI Assessment in the top navigation bar to open the chat-based intake. Describe the company you are assessing in the chat, and the platform auto-researches the company and fills a Business Requirements panel through natural conversation.

Click Initiate assessment. A dialog asks which route to take:

• Yes - full AI Readiness (~12 min) - adaptive Q&A across 6 categories, use cases grounded in your answers, plus 6 scored PDF reports (Revised Priority Use Cases, Platform Decision, Gap Mitigation Plan, Feasibility, ROI Modeling, Well-Architected Review)
• No - use cases only (~8 min) - use cases generated directly, no questions, no extra reports

Business email is required. Personal email domains are blocked at signup and at the Use Case Tool form.

On the full route, six reports tell you what is blocking you. See Reading Your Readiness Reports.

Prioritising use cases

Both assessment routes produce a prioritized use case list and land on the Jamming Board - a collaborative workspace where your team rates use cases with star ratings, adds threaded comments, and ranks them automatically via combined rating and sentiment analysis. The top-ranked use case gets a Best Fit badge. Custom use cases can be added manually. Additional artifacts (Executive Presentation, AWS Architecture Diagram) can be generated on demand.

What you leave with: a readiness picture and one use case selected to build. The ranking is an input to that decision, not the decision itself - see Choosing Your First Use Case.

Align

Before designing agents, confirm the work is a fit and the foundations exist.

Execution pattern. Agentic Process Transformation reads your IT application inventory and proposes which execution pattern each application's work should take - Deterministic, Agent as Tool, or Other. Work that is genuinely deterministic does not need an agent.

Data and knowledge. Enrich your agents with domain knowledge before or after deployment:

• Knowledge Base - upload documents (PDF, DOCX, and more) for unstructured RAG
• Structured Dataset - connect tabular data sources
• Integrations - configure MCP tool connections
• GraphRAG - build a domain property knowledge graph via the AI-powered graph builder and Cytoscape.js canvas. The GraphRagAgent extracts entities and relationships, grounds each node in document evidence, and stores the graph as a versioned S3 artifact. Agents query it at runtime for precise relationship reasoning.

Boundaries. Access model, compliance evidence, and security posture are set in Governance & Guardrails.

What you leave with: the data, connections, and boundaries this use case needs. These are scoped to this project - a later use case is a separate project and sets up its own.

Launch

From the Jamming Board, click Design App on your chosen use case to create an Application and enter the Control Hub.

Agentic Workflows

The Workflow Architect generates multi-step workflow plans tailored to your use case. Each workflow has a name, description, and ordered step list. Limits: 30 workflows per project, 300 KB total, 24 KB per workflow. Remaining capacity is shown before you hit the cap.

Agentic System

The Workflow Architect runs in agent-design mode to generate a multi-agent architecture - typically 3-7 specialized agents, each with a defined role, model, tools, memory, and system prompt. You can edit any agent inline or refine the entire system via the chat panel in plain English.

Deployment to AWS Bedrock AgentCore fires automatically once the spec is finalized - there is no manual Approve step. Deploy logs stream in real-time.

Test Bed

A live chat environment connected to your deployed AgentCore runtime. The Agent Activity Tree shows real-time sub-agent orchestration - each agent's status, tool calls, reasoning, and responses. Run structured Evaluations across up to 8 metrics spanning Response Quality, Task Completion, Tool Usage, and Safety.

Deploy

Three inner tabs:

• Generate App - add optional instructions and a Brand Kit (name, tagline, color, font, logo), then click Generate App. Code generation streams in real-time. Generated files appear as NEW or MOD cards with full source. On completion, code pushes to GitHub and Amplify builds and deploys it automatically. A Live Deployment card appears with your live URL.
• API - API endpoint URL, key management (max 2 per project), Quick Start code, and Developer Guide generation via the DevGuide agent.
• Business Case - cost estimates across Conservative/Base/Growth scenarios, AWS Architecture Diagram, and PDF export.

What you leave with: a live application at its own URL, with agents running on AgentCore.

Operate

Once your app is deployed, the Application Management dashboard provides ongoing control across ten tabs:

What you leave with: evidence about what building one of these actually costs - which makes your next use case selection better informed than the first.

Closing the Loop

Operate returns to Envision. The ranked list from your assessment is still on that project's Jamming Board, and you can now re-read it against what you learned rather than against the original estimate.

The Fast Path

You can skip Envision and Align entirely by selecting a Blueprint. Blueprints are pre-built, production-ready multi-agent systems for 17 enterprise use cases across five categories:

Complete the setup wizard (2-3 minutes), and Qubitz customizes and deploys the blueprint before dropping you into Control Hub to review agents and generate the app - landing you directly in Launch.