Trusty AI: Overview
Ask Trusty questions on your metadata.
Trusty AI is built as an AI-powered semantic layer for your metadata. It reduces "search friction" by allowing you to identify relevant data assets using natural language instead of manual browsing or strict naming conventions.
Core Capabilities
Semantic Discovery: Find datasets across warehouses using business metrics or plain language.
Lineage Analysis: Map impact pathways and understand data dependencies through simple queries.
Operational Health: Synthesize monitor statuses and incident history to understand data reliability.
Data Profiling: Read column-level statistics — null rates, value ranges, cardinality, and distributions — directly from Decube's profiling data. Compare historical profiles to identify trends or regressions over time, and get recommendations for which monitors to set up based on profiling signals.
To see a list of prompts that you can ask TrustyAI, go to Suggested Questions to ask Trusty
Skills
Trusty AI can also run skills — prebuilt, repeatable expert workflows such as Monitor Recommendation. The same skills are available for download into your own AI client. For the full list and details, see Skills.
AI Agent Registry
Trusty sits in the AI Hub alongside the AI Agent Registry, an inventory of the AI agents running against your data.
AI Agent Registry: OverviewWorking with responses
Copy anything: Copy or download a single code block (SQL, Python, JSON) or an entire response in one click.
Work with tables: Copy or download a table as CSV or Markdown, or maximise it into an expanded view for easier reading.
Visualise: Trusty detects when an answer is best expressed visually and renders it inline in the conversation. Use cases such as but not limited to lineage, pipelines, and relationships can be rendered as diagrams that you can copy and download.
Regenerate: Re-run your question for a fresh answer. Trusty adds the new response below the original, so you keep both.
Accessing Trusty AI
Trusty AI is currently in a BETA phase. If it is enabled for your organization, follow these steps to access it:
Log in to your Decube instance.
Click AI Hub in the left navigation bar. The AI Hub opens on the Trusty chat by default.
Type your request or question into the chat interface at the bottom of the screen.
If you do not see AI Hub in your navigation bar, please contact your Account Manager to request access.
Security and Data Usage
Trusty AI is built with enterprise-grade security as a primary requirement:
Metadata Only: Trusty AI interacts exclusively with extracted metadata. It cannot execute queries against your raw data sources and has no permission to view record-level data.
Copying and downloads stay within your permissions: Copy and download actions only ever contain contents you are already authorised to see under RBAC. Once you download or copy content, it sits outside Decube's controls.
Infrastructure: The assistant utilizes AWS Bedrock foundation models.
For SaaS customers, this is managed within Decube’s secure AWS environment.
For BYOC (AWS) customers, Trusty connects via your own AWS Bedrock API to ensure data stays within your infrastructure.
Data Retention: Decube stores natural language prompts for review and troubleshooting purposes only.
Usage and Quotas
To ensure system stability, quota limits are applied to every account.
You can view your remaining quota directly below the search bar in the Trusty AI interface.
Regenerating a response re-runs your question against the model and counts toward your quota.
FAQ
Does Trusty AI have access to my actual data records?
No. Trusty AI only accesses the metadata (table names, column names, descriptions, and profiling statistics) already ingested by Decube. It never sees the individual rows of data in your warehouse.
What happens if Trusty AI provides an incorrect answer?
As Trusty AI is in a beta phase, it may occasionally generate "hallucinations" or factually incorrect statements. Verify critical information (like lineage dependencies) via the Decube UI. Please report these errors to support@decube.io.
Can Trusty AI make changes in Decube?
Only one: it can create a monitor. Trusty proposes a monitor with a recommended test type and configuration, and creates it only after you approve, and only if your RBAC permissions allow you to create monitors. Every monitor created this way is attributed to you in Decube's audit trail. Everything else is read-only: Trusty cannot update descriptions, change ownership, or modify any other asset or documentation within Decube as of now.
Why can't I access Trusty AI on my BYOC Azure environment?
Currently, Trusty AI only supports integration via the AWS Bedrock API. Customers on Azure-based self-hosted deployments are not yet eligible for the current phase Trusty AI beta. If you are hosted on Azure and would like to be on the waitlist for this, please express interest to your Account Manager.
Who can see my chat history?
Chat histories are visible to you to help you revisit previous sessions. However, Decube team may review anonymized logs strictly for troubleshooting and model optimization purposes.
Last updated