Quick Start

Navigate to qubitz.ai and click Login.

Enter your email and password. After authentication, you will be redirected to the App Builder.

Type a description of what you want to build in the input area and press Build my solution. As you describe your problem, the left panel tracks your progress under What we'll build:

• PROJECT NAME - derived from your description
• SHAPE THE PROBLEM - five signals the platform collects to frame your build:
  • Workflows to automate
  • Who uses it
  • Data sources / systems
  • Integrations needed
  • What success looks like
• OPTIONAL - additional context you can provide to enrich the output

The more detail you give, the better the generated agent system will be. Once you hit Build my solution, Qubitz routes you into the right flow based on your input.

After App Builder processes your input, you land in Control Hub with your project already named and your workflows generated. The left panel gives you three sections:

• Agentic Workflows - your business workflows, each as a runnable card
• Agentic System - the multi-agent architecture behind those workflows
• Data - domain knowledge sources your agents can query at runtime

Agentic Workflows opens by default and shows a card for every workflow generated under Business Workflows:

• Each card shows the workflow title and a one-line description of its deliverable
• Hit Run to trigger a workflow
• Drag the handle on the left to reorder
• The New workflow card at the bottom lets you describe a new workflow in plain English and the platform designs it for you

Agentic System shows the numbered list of AI agents the platform built:

• Each card shows the agent name and its role description
• Expand any agent to edit its model, tools, memory, visibility, system prompt, and agent-as-a-tool configuration
• Add a custom agent card at the bottom lets you define a new agent with its own prompt, model, tools, and memory
• Platform Agents section below gives you access to shared system-level agents
• The right panel - Change with AI - lets you refine the entire agent system in plain English: describe what should be different and the platform rewrites the relevant agents, tools, and workflows

The Data tab has four sub-tabs for enriching your agents with domain knowledge.

Knowledge base - Documents your agents read and quote back, with citations. Contracts, policies, reports, manuals.

• Collections - an isolated set of documents; each agent only searches the collections you bind it to. Create as many collections as you need and assign them per agent
• Files - upload documents directly into a collection
• Data sources - connect a repository and pick which documents to index; only what you select is indexed. Supported sources:
  • Amazon S3, SharePoint, OneDrive, Google Drive, Confluence, Box, Dropbox, File share (SMB), Web crawler, Custom source

Structured data - Tables your agents count, filter, and total over for exact numbers. Spreadsheets and databases.

• Files - upload a spreadsheet (CSV, XLSX) for agents to query
• Data sources - connect a live database read-only; nothing is uploaded or copied, and you choose which tables agents can see. Supported databases:
  • PostgreSQL, MySQL / MariaDB, Amazon Redshift, Microsoft SQL Server, Oracle, Snowflake, Google BigQuery

Connections - Apps and tools your agents can act in, not just read. Connect once, then choose which agents may use each connection.

• Connect in one click - Slack, Gmail, Google Calendar, GitHub, GitLab
• More apps (OAuth sign-in, two-step setup) - Experian, Atlassian, Salesforce, ServiceNow, HubSpot, Zendesk, Notion, Linear, Sentry

Graph RAG - Relationships across your documents so agents can follow how entities connect, not just match keywords.

• Describe what you want the knowledge map to answer, then click Build my map
• The platform extracts entities (nodes) and relationships (edges) from your collections and builds an interactive graph
• Three build steps: Node graph - Ontology - Pipeline
• Use Advanced: guide the build to control entity types, relationship rules, and extraction behaviour before building

From the Agentic Workflows tab, click Run on any workflow card. Fill in the inputs and submit.

Running the workflow

• Each agent is called in the order defined by the workflow steps
• Watch each agent being invoked, the tools it calls, and the output it produces in real time
• Results appear inline as each step completes

Reviewing the output

Once the run is complete, the output is presented inline as a structured summary of every stage with its status and key findings. Any documents generated during the run appear as downloadable files in the Agent Activity panel on the right.

Evaluate

Click Evaluate to score the run. Select up to 8 metrics across four dimensions:

• Response Quality - Correctness, Helpfulness, Coherence, Conciseness, Faithfulness, Relevance, Instruction Following, Refusal Detection, Context Relevance
• Task Completion - Goal Success
• Tool Usage - Tool Selection, Tool Parameters
• Safety - Harmfulness, Stereotyping

Click Run Evaluation and the platform returns an Evaluation Results summary with an overall percentage score, a rating per metric, and a detailed explanation for each one.

Safety & Stress Test

Click Safety & stress test to run adversarial, multi-turn attacks against your live agent - separate from quality scoring. The platform automatically attacks your agent with known jailbreak techniques to see if it can be tricked into fabricating data, ignoring its rules, or leaking its instructions. Each technique is tested against 3 threats. Pick up to 3 attack techniques:

• Gradual escalation (Crescendo) - Starts with an innocent question and slowly steers the conversation toward the goal over several turns
• Iterative refinement (PAIR) - An attacker AI keeps rewriting its prompt based on the agent's replies until it finds one that works
• Adaptive attacker (GOAT) - An attacker AI improvises, mixing different jailbreak tricks on the fly instead of one fixed script
• Rating-scale trick (BadLikertJudge) - Tricks the agent into acting as a rating judge, then uses that framing to extract content it would normally refuse
• Step-by-step break (SequentialBreak) - Splits the request into small, individually harmless steps that together assemble the forbidden output

While the test is running, a live log streams every attack attempt in real time - which technique is active, which attack number is in progress, and what the agent is being tested against.

Once all attacks complete, you can review the results to see which techniques succeeded, which were blocked, and where your agent held its ground.

When you are ready to deploy, click the Deploy button in the Agentic Workflows tab. A dialog opens asking where you want your agentic system to run:

• Qubitz Cloud (Fastest) - Deploys into the Qubitz AWS account. Nothing to set up - it deploys straight away and you land in your workspace
• Your AWS account (BYOC) - Brings your own cloud. The solution is deployed into your AWS account so your data and runtime stay with you. Takes about five minutes of setup

Before committing, click Business case to review the cost estimates and AWS architecture for your use case. The Business Case has three tabs:

Cost Estimates - shows your projected Monthly MRR, Annual ARR, and Buffer percentage, followed by a full service-level breakdown of every AWS service your deployment uses - each with a description of what it does and its monthly cost.

Architecture - renders a full AWS Solution Architecture diagram for your use case, covering your multi-agent system, the Bedrock AgentCore runtime, security and governance, observability, and the deployed customer application layer. Use Open to view it full-screen or Regenerate to rebuild it.

Export - two options once you are ready:

• PDF Report - export the full business case as a PDF
• AWS Calculator - open the AWS Pricing Calculator to verify costs independently

Once deployed, your project opens in the Application Management dashboard - a dedicated workspace for everything that runs in production. The sidebar is split into three sections: Foundation, Application, and Operations.

Foundation

The Foundation section covers what your product is built on - the workflows it runs, the agents behind them, and the data they work from.

Business Workflow

The landing tab. Shows every workflow your agents can run as cards under Business Workflows:

• Each card displays the workflow title, a one-line summary of its deliverable, and a Run button to trigger it
• Drag the handle on the left to reorder workflows
• The New workflow card at the bottom lets you describe what you want to build and your agents will design it

Agentic System

The same agent editor as Control Hub - your full agent roster is editable from here without going back to design mode:

• Numbered agent cards, each expandable to edit model, tools, memory, and system prompt
• Guardrails button (top right) to configure content safety rules across all agents
• Add a custom agent card at the bottom to define a new agent with its own prompt, model, tools, and memory
• Platform Agents section below for system-level agents managed by Qubitz
• Right panel has two modes:
  • Chat - describe what should be different and the panel rewrites the agents, their instructions, and the tools they can reach
  • Logs - view real-time execution logs from the agent runtime

Data

Manage the domain knowledge and data sources your agents query at runtime. Four sub-tabs:

• Knowledge Base - documents and files your agents reference during a run; organised into named collections, each bound to specific agents
• Structured Data - CSV files and database connections your agents can count, filter, and total over; datasets are grouped into collections and each collection is bound to the agents that need it
• Connections - third-party integrations linked to this project (CRMs, APIs, MCP servers)
• Graph RAG - your domain property knowledge graph built from uploaded documents and queryable as a structured agent tool at runtime

Application

The Application section covers the deployed customer-facing web app - building it, configuring it, and controlling who gets in.

App Generator

Build and deploy the web app your users will actually open:

• Source repository - the GitHub repo Qubitz created for this project is linked here
• Generate the application - customise your brand kit, then click Generate app to write, build, and deploy the frontend; Amplify picks up the push and compiles the live app automatically
• Business Case - opens the cost estimate and architecture diagram for this deployment

App Config

Configuration for the live app and its API:

• Application Configuration - add a GitHub collaborator and set your AWS Bedrock API key so inference bills to your own account
• Custom Domains - add and verify a custom domain for your deployed application (available once app is generated)
• API Endpoint - your agent's WebSocket endpoint URL for external integrations; copy it directly from this panel
• API Keys - generate and manage API keys for programmatic access to your agent (up to 2 keys per project)
• Channel Integration - put your agent where your users already are: Web Widget, Slack, WhatsApp, Microsoft Teams, and Voice via Twilio and Amazon Nova Sonic
• Build History - every previous Amplify build with version number, status, live URL, and timestamps

Identity & Access

Controls who can reach your agent and what they can do:

• What is set up - shows the current trust model (who is asking) and access rules (what everyone gets by default)
• How the assistant knows who is asking - choose between two modes:
  • Trust whoever calls - the caller declares the user identity and the agent believes it; simplest mode, right when only your own app can reach the agent
  • Check every caller - the agent verifies the caller's sign-in token against their identity provider before responding; nobody can impersonate another user
• What each group can reach - define named groups and set exactly which capabilities each group can access; groups sync from your directory or are added manually
• SSO Connections - connect Microsoft Entra ID, Google Workspace, Okta, Auth0, or any SAML/OIDC provider
• Directory Sync (SCIM) - automatically provision and deprovision users from your corporate directory
• Domain Verification - verify the email domains your users sign in from
• Sign-in Policy - set MFA requirements and session timeout rules
• Grants - assign scoped access grants to users and agents within this application
• Agent Auth - configure service principal credentials so agents can authenticate as trusted identities when calling external systems

Operations

The Operations section is for monitoring what is already running in production - traces, compliance evidence, and spend.

Observability

Live runtime telemetry for your deployed agent, pulled automatically from AWS CloudWatch and X-Ray:

• Five summary metric cards at the top: Total Sessions, New Sessions, Total Traces, Avg Latency (in seconds), Error Rate
• Token Usage by Model - total tokens consumed per model with call count, input token count, and output token count broken out
• Resource Consumption - vCPU hours, memory (GB-hours), and throttle rate for the agent runtime
• Error Breakdown - error rate percentage, system error count, and client error count
• Agent Spans - per-agent list showing trace count and average latency; every sub-agent and tool call is tracked individually so you can see exactly which step is slow
• Metrics refresh automatically every 60 seconds; click Refresh to pull the latest data immediately

Compliance & Security

Security testing and compliance evidence for this application. Four sub-tabs:

• Security - on-demand penetration testing powered by AWS Security Agent; click Run pen test to start; findings, evidence, and a downloadable PDF report are saved automatically and ready to share with your security or audit team
• Compliance - DPO Pack: seven GDPR compliance documents generated in parallel (Privacy Policy, Data Flow Map, Security Policy, Retention Policy, Subprocessor Register, Cookie Policy, GDPR Controls)
• WAFR - AWS Well-Architected Framework Review scored across six pillars: Operational Excellence, Security, Reliability, Performance Efficiency, Cost Optimization, and Sustainability
• Design Document - full application design document covering architecture, data models, API contracts, and implementation details

FinOps

AWS cost data for this application with three period views - This Month, Last 7 Days, Today:

• Three summary cards: Month to Date total spend, Services count, Daily Average spend
• Cost Breakdown by Service - itemized cost per AWS service; Amazon Bedrock costs are shown as estimated when you are on Qubitz tokens rather than your own Bedrock key
• Daily Cost Trend - spend per day across the selected period as a horizontal bar chart
• Cost data can take up to 24 hours to appear after first deployment; not available for BYOC deployments

Settings

Project-level configuration for the Qubitz workspace:

• Application Configuration - Application ID, environment (Production), and AWS region (e.g. eu-central-1 / Frankfurt)
• Access Control - shows every user with access to this project and their role (Admin, Operator, Viewer)
• Blueprint Catalogue - click Save as Blueprint to share this application's full configuration as a reusable blueprint for others on the platform
• Revert from Production - moves the project back to draft so you can edit and redeploy from Control Hub again; the app stays deployed while you make changes
• Collaborators - invite team members by email and manage their access; roles appear in the Access Control panel once accepted
• Danger Zone - Teardown permanently deletes this application and every AWS resource it created, including the agent runtime and storage; this cannot be undone