AI Copilot
Copilot Overview
The StackFlow AI Copilot is a persistent AI assistant sidebar available on every page in the portal. Unlike a generic chatbot, the Copilot is fully context-aware — it knows which record you're viewing, your role, your organization's knowledge base content, and the history of your current session. This context enables highly specific, actionable answers rather than generic guidance.
- Bedrock KB:
BXJGG7PIPSactive for copilot knowledge retrieval - DynamoDB:
StackFlow_TenantAIConfigmust havecopilotEnabled: truefor the tenant - Redis: Copilot session state stored under
sf:session:{sessionId}with TTL 1800s - Feature Flag:
ai_copilotflag enabled inStackFlow_FeatureFlag
The Copilot uses a RAG pipeline backed by the Bedrock Knowledge Base (BXJGG7PIPS) for grounded answers. It also has direct read access to the current ITSM record's data, allowing it to reference specific field values in its responses. For example, when viewing a P1 incident, the Copilot can analyze the specific error description and suggest targeted resolution steps.
In-Context Awareness
The Copilot automatically receives context from the current page:
| Page | Context Provided to Copilot |
|---|---|
| Incident detail | Short description, description, state, priority, CI, resolution notes |
| Change detail | Change description, risk score, affected CIs, approval history |
| CMDB CI detail | CI attributes, relationship graph summary, incident history |
| Knowledge article | Article title, content, category, related records |
| Dashboard | Visible metric values, active alerts |
Copilot Actions
The Copilot can take direct actions in StackFlow on behalf of the user (with confirmation). Available actions include: create a work note on the current record, add a tag, update the priority, create a linked knowledge article, trigger an AI workflow, or run an impact analysis query. Action buttons appear in the Copilot response when an action is recommended.
Customizing the Copilot
Administrators can customize the Copilot's system prompt template (copilot_system) to inject organization-specific instructions, tone guidelines, and domain knowledge. Common customizations include: specifying the organization's IT jargon and acronyms, providing escalation contact information, and instructing the Copilot to always check a specific internal wiki before answering.
Copilot for Agents vs End Users
The Copilot behaves differently based on the user's role. For ITSM agents, it provides technical troubleshooting guidance, diagnostic commands, and internal knowledge. For end users (via the self-service portal), it focuses on guided self-service, explains where to find help, and avoids exposing internal technical details. This behavioral split is controlled by the copilot_role_mode prompt variable injected by the model router.