AI Agents - Technical Specs
AI Models & Infrastructure
Large Language Models
Monte Carlo uses Anthropic's Claude models and Amazon Titan embeddings accessed via Amazon Web Services (AWS) Bedrock and customer data is never used for model improvement.
Data Retention
Amazon Bedrock operates under a zero data retention (ZDR) security model. By default, Bedrock does not store the inputs (prompts) or outputs of the models Monte Carlo uses; data is processed in-region only to serve the request and is not retained. Every model listed below is covered by this default ZDR behavior, and Monte Carlo does not use any Bedrock model that carries extended abuse-monitoring data retention.
Model Availability by Region
Not all models are available in all AWS regions. The table below shows which model versions are used for Monte Carlo AI features based on your instance region:
| Model | US | EU | APAC | Model or System Documentation |
|---|---|---|---|---|
| Claude Sonnet 5 | ✓ | ✓ | ✓ | Anthropic |
| Claude Sonnet 4.6 | ✓ | ✓ | ✓ | Anthropic |
| Claude Sonnet 4.5 | ✓ | ✓ | ✓ | Anthropic |
| Claude Sonnet 4 | ✓ | N/A | N/A | Anthropic |
| Claude Opus 4.8 | ✓ | N/A | N/A | Anthropic |
| Claude Opus 4.6 | ✓ | N/A | N/A | Anthropic |
| Claude Haiku 4.5 | ✓ | ✓ | ✓ | Anthropic |
| Claude Haiku 3 | N/A | ✓ | ✓ | Anthropic |
| Amazon Titan Text Embeddings V2 | ✓ | ✓ | ✓ | AWS |
Model versions may be updated as newer Claude models become available in Bedrock. If for some reason the above links are unavailable, this documentation is typically located on Anthropic's Trust Center.
System Requirements
Optimal Requirements
For best AI feature performance:
- Data sampling enabled: Required for some features, enhances others
- Query logs enabled: Provides usage patterns and lineage data for AI analysis
- Full lineage instrumentation: Connections to upstream/downstream systems
- Active integrations: GitHub/GitLab, dbt, Airflow, or other orchestration tools
- Historical data: At least 7-14 days of monitoring data for pattern recognition
Features That Require Data Sampling
Some AI capabilities require data sampling to be enabled:
- Monitoring recommendations that analyze field values
- Data pattern detection and anomaly identification
- Features that assess data quality at the record level
When data sampling is disabled, these features become unavailable, while metadata-only features continue working.
Supported Use Cases
AI features are designed for:
- Data quality monitoring: Recommending monitors, detecting anomalies, assessing coverage gaps
- Incident investigation: Root cause analysis, hypothesis testing, correlating changes
- Query assistance: SQL generation, query explanation, optimization suggestions
- Conversational troubleshooting: Follow-up questions, iterative refinement, context retention
Updated 12 days ago