Knowledges is Castrel's operational knowledge management system that stores your team's operational experience, troubleshooting guides, and system documentation in a structured format. This knowledge is automatically retrieved and referenced by Castrel AI, helping it better understand your system environment and provide more accurate diagnostics and recommendations.
Unlike traditional documentation, Knowledges uses a Trigger mechanism that activates knowledge at the right moment. For example, when the AI analyzes an alert for a specific service, knowledge bound to that service is automatically included in the context.
Knowledge Types
Castrel supports two knowledge types:
| Type | Description | Use Cases |
|---|---|---|
| Simple | Plain text content | Brief configuration notes, caveats, quick references |
| Document | Markdown-formatted documents | Detailed architecture docs, troubleshooting guides, best practice manuals |
Trigger Conditions
Trigger conditions determine when knowledge is referenced by the AI. Choosing the right trigger ensures knowledge is activated in the most relevant scenarios:
| Trigger | Description | Example |
|---|---|---|
| Always | Always active, available in all conversations | Company-wide ops policies, general security guidelines |
| Application | Bound to a specific application | Payment app deployment procedures |
| Service | Bound to a specific service | Order service architecture documentation |
| Infrastructure | Bound to specific infrastructure | MySQL primary database configuration guide |
| Connector | Bound to a specific data connector | Prometheus query tips and tricks |
| Alert | Matches based on alert content | "OOMKilled" troubleshooting runbook |
Manual Creation
In the Castrel console, navigate to Settings > Knowledges and click Create Knowledge:
payment-gateway-architecture)Creating from Conversations
When Castrel AI generates valuable analysis reports during troubleshooting, you can save them as knowledge with one click:
This approach is ideal for capturing troubleshooting experience, turning your team's incident history into reusable knowledge assets.
Creating from Resource Detail Pages
In the detail pages for Applications, Services, Infrastructure, and Connectors, you'll find a Related Knowledge tab:
When properly configured, knowledge automatically takes effect in these scenarios:
Alert Analysis
User: Analyze the high latency alert for mysql-prod-01
AI: I've retrieved knowledge documents related to mysql-prod-01...
Based on the "Database High Latency Analysis Report", similar issues
are typically caused by:
1. Missing indexes leading to full table scans
2. Slow query backlog
3. Connection pool misconfiguration
Let me start investigating...
Incident Investigation
When diagnosing incidents, the AI automatically references:
Q&A Queries
User: What's the timeout threshold for our payment service?
AI: According to the "Payment Gateway Architecture" document,
the payment service timeout settings are:
- Default timeout: 30s
- Retry attempts: 3
- Concurrency limit: 1000 req/s
| Tip | Good Practice | Poor Practice |
|---|---|---|
| Precise Binding | Bind knowledge to specific services or infrastructure | Set all knowledge to Always active |
| Structured Content | Use Markdown headings, lists, and code blocks | Large blocks of unformatted text |
| Keep Updated | Update knowledge after system changes | Knowledge inconsistent with actual config |
| Use Labels | Add labels like env=production, type=troubleshooting | No classification labels |
| Alert Matching | Use precise keywords like "OOMKilled", "Connection refused" | Overly broad matching terms |
Ideal For:
Not Ideal For: