Workspace
Conversation
Demo Guide How to explore this demo
You’re interviewing employees at Northstar, a fictional industrial company, to understand how a business process really works. The agent can investigate connected systems and company documentation while building a structured process model.
Demo environment: CRM, pricing and company documents contain synthetic data.
Describe how work actually happens — including manual steps, delays, exceptions, approvals and workarounds.
Recommended journey
Suggestions fill the composer onlyWhat you’re watching
Plain-English architecture- Conversation
- The GPT model interviews the SME and decides the next action.
- Process Model
- Application-owned structured knowledge extracted from discovery.
- MCP
- Interact with structured business systems and tools.
- Knowledge / RAG
- Search unstructured company knowledge semantically.
- Analysis
- Application-controlled Analyst → Designer → Verifier orchestration.
- Evaluation
- Offline behavioral regression testing of the AI system.
Keep an eye on Agent Activity to see whether a turn uses the LLM alone, MCP, RAG, state extraction or orchestration.
Once enough process knowledge is captured, run Analyze discovered process to see Analyst → Designer → Verifier orchestration.
Architecture
- MCP client → Northstar MCP server → CRM / Pricing
- RAG → embedding model → local vector index → company documents
- ProcessState → application-owned structured process knowledge
Interview the Northstar team
Describe a messy process as it really works. The agent will ask questions and the execution trace will show what happens underneath.