#
Core Concepts
#
Architecture
EGPT orchestrates three distinct resource types:
- Digital Experts (Skills): The Action Layer. Use specific, pre-built logic modules for tasks (e.g., risk calculators, code debuggers).
- Enterprise Knowledge Gardens (EKGs): The Knowledge Layer. Access secure repositories of proprietary data (e.g., HR policies, product specs).
- External Connectors: The Existing Layer. Integrate real-time data sources (e.g., Web Search) for current context.
- EGPT Engine: The Unified Conversational Screen. Acts as the central intelligence layer managing conversation flow.
#
Operational Modes
EGPT supports users through three modes:
- Assist Mode: The High-Speed Knowledge Navigator. Instant answers referenced from EKGs (e.g., finding HR policies).
- Augment Mode: The Co-pilot for Complex Decisions. Combines Skills and EKGs for deeper analysis (e.g., comparing vendor contracts against risk policies).
- Automate Mode: The Engine for Process Execution. Triggers pre-defined system actions from a prompt (e.g., generating IT tickets).
#
Governance & Observability
The Purple Fabric Governance Module provides full visibility into EGPT usage. Metrics are visible only to tenant admins and include:
- Adoption Trends: Total Messages, Active Users.
- Cost & Efficiency: Total Model Cost, Latency.
- Token Utilization: Input vs. Completion tokens.
- System Reliability: Transaction status breakdowns.
#
EGPT Resource Promotion and Visibility
There are two primary levels of visibility for a resource - Private and Public.
- Private Resource
When a resource (agent or EKG) is designated as Private, it can be promoted to EGPT but will only be visible and accessible to its creator. This allows for the use of personal assets within the EGPT framework without broader sharing. - Public Resource
A Public resource is designed for wider distribution. When a creator promotes a public resource, it becomes available for their personal use within EGPT. Broader visibility and access are controlled by authorized administrators.
Visibility:
A Public resource is visible to its creator, as well as the Workspace Admin and the Tenant Admin.
Access Control:- A Workspace Admin can make the public resource available across their entire workspace.
- A Tenant Admin can extend this access, making the public resource available across the entire tenant, allowing users in any workspace to use it.
#
Compatibility Matrix
This table provides a quick overview of who has visibility and access to a resource based on its promotion status.
#
Agent Promotion Scenario
Imagine Sarah, a financial analyst at a major bank. She's tired of manually generating quarterly performance reports. To save time, she builds an EGPT Agent that can automatically pull data, analyze trends, and create a summary report.
1. Private Promotion
Sarah first promotes this agent as a Private Resource for her own use.
Creator (Sarah) -> Private Agent -> Visible only to Sarah -> Promoted to EGPT
2. Public Promotion
Once she knows it's a great tool, she promotes it as a Public Resource.
Creator (Sarah) -> Public Agent
What happens now? Her Workspace Admin (the head of her department) sees the agent in the EGPT library. Recognizing how useful it is, the admin enables it for the entire financial analysis team.
Workspace Admin -> Enables for Team -> Visible to everyone in Sarah's team
Later, the Tenant Admin (the head of all company workspaces) sees the agent. They realize it could benefit all financial teams, not just Sarah's. The Tenant Admin then promotes it to the entire organization.
Tenant Admin -> Enables for Company -> Visible to all users in the organization
#
Observability & Governance
#
Monitoring & Metrics
Purpose
Provides tenant admins and Workspace Admin with comprehensive visibility into EGPT performance, usage trends, and costs through standardized governance dashboards. Ensures EGPT adoption aligns with enterprise efficiency, compliance, and business value objectives.
Access Control
- Admin Only – EGPT metrics are visible exclusively to tenant admins in the Governance module.
- Non-Admin Users – No access or visibility to EGPT metrics.
Standard Metrics (aligned with Conversational Agents)
- Total Messages
- Total Model Cost (includes both file processing and main agent operations)
- Input Tokens / Completion Tokens
- Average Model Latency (ms)
- User Satisfaction Score (if recorded)
Graphs Available
- Messages Over Time
- Users Over Time
- Model Cost Over Time
- Token Usage Over Time (Input/Completion toggle)
- Latency Over Time
- Transaction Status Breakdown (Initiated, In Progress, Completed, Failed)
Message Log Table Columns
- Message ID
- Status
- Input Tokens
- Completion Tokens
- Model Cost
- Model Latency
- Initiated By
- Initiated On
- Duration
- User Satisfaction
Message-Level Drilldown
- Input/output prompts
- Tool/agent execution sequence
- Token usage and latency
- Total cost
- Decision logic (if available)
Traceability
- All message-level data is captured and retained for audit compliance, regardless of UI visibility.