RAG & Knowledge Base
AtlasAI uses Retrieval-Augmented Generation (RAG) to ground its AI reasoning in your organization’s operational history. Rather than relying solely on pre-trained model knowledge, the AI retrieves relevant context from your past incidents, runbooks, documentation, and configurations before generating responses.
What Gets Indexed
The RAG knowledge base automatically indexes:
| Source | Content | Update Frequency |
|---|---|---|
| Incidents | Titles, descriptions, RCA results, resolution summaries, post-mortems | On incident close |
| Runbooks | Step definitions, execution logs, success/failure outcomes | On save/execute |
| CMDB | Service descriptions, relationships, configuration attributes | On change |
| Documentation | Internal wiki pages, architecture docs, operational procedures | On sync |
| Chat history | Copilot conversations flagged as “useful” by operators | On flag |
| Change records | Change descriptions, implementation plans, rollback procedures | On change close |
How RAG Enhances AI
During RCA
When analyzing an incident, the RCA engine queries the knowledge base for:
- Past incidents affecting the same service or similar services
- Incidents with matching error patterns or metric signatures
- Known failure modes documented in architecture docs
- Recent changes that could explain the current behavior
This context is injected into the reasoning prompt, allowing the AI to say things like: “This pattern matches INC-00027 from two weeks ago, which was caused by a database connection pool exhaustion after a deployment.”
During Runbook Generation
The AI retrieves:
- Runbooks that previously resolved similar root causes
- Operator-edited runbooks that refined AI-generated steps
- Service-specific operational procedures from documentation
During Copilot Conversations
The Copilot retrieves:
- Relevant incident history when you ask about a service
- Operational procedures when you ask “how do I…”
- Configuration details from the CMDB when you ask about infrastructure
Knowledge Base Management
Viewing Indexed Content
Navigate to Settings → AI → Knowledge Base to see what’s indexed:
- Total document count and index size
- Recent indexing activity
- Search the knowledge base directly to verify content
Manual Content Import
You can manually add content to the knowledge base:
- Navigate to Settings → AI → Knowledge Base → Import
- Upload Markdown, PDF, or plain text files
- Add metadata tags (service, team, category) for better retrieval
- Content is vectorized and available for retrieval within minutes
Data Retention
Knowledge base entries follow your configured data retention policy. By default:
- Incident data is retained for 2 years
- Runbook data is retained indefinitely
- Documentation is retained until manually removed
- Chat history is retained for 90 days
Configure retention under Settings → Data Management → Retention Policies.