Data Handling
Telmai's data handling model rests on one distinction. The platform processes your data where it lives and stores only what it derives from it. Raw records are read to compute quality results and are never copied into long-term storage in Telmai's infrastructure.
What Telmai Processes And What Telmai Stores
Raw customer records
Table rows, file contents, field values
No. Read during scans to compute results, never retained
Derived quality results
Aggregate metrics, profiling statistics, drift measurements, data quality scores
Yes
Schema and asset metadata
Attribute names, types, nullability, key flags, glossary terms, asset ownership
Yes
Operational records
Monitors, alerts, incidents, job history, user actions
Yes
Connection credentials
OAuth tokens, integration credentials
Yes, encrypted with AES-256-GCM
Everything Telmai stores is derived or operational. None of it contains your record-level values.
Where Your Data Is Processed
Where processing and storage happen depends on your deployment model.
Customer VPC - The entire platform, including scanning, derived data storage, and logs, runs inside your cloud account under your network, IAM, encryption key, and residency controls. Nothing leaves your environment.
SaaS - Scans read your data through the connection you configure, and derived results are stored in Telmai's cloud infrastructure. In evaluation and POC environments, profiling runs on physically separated compute per customer, and customer data is purged once metrics are computed.
Microsoft Fabric workload - Data quality computations execute inside your Fabric engine through query push-down, and only derived data reaches Telmai's infrastructure. Regional details are stated in Telmai's Data Residency Commitments, the canonical residency reference.
Third Parties
Customer data reaches third parties through exactly two paths. Cloud provider services that Telmai's infrastructure runs on, and integrations you explicitly configure, such as Jira, Slack, Teams, email, or your own webhook. In both cases, what moves is derived data and notifications, never raw table rows.
AI model providers are a special case with their own strict boundary. Prompts carry schema metadata, operational metadata, and derived aggregates only, and never record-level values, personal data, or credentials. How that boundary is enforced is covered in AI Architecture and Data Handling.
Logs And Retention
Application logs support diagnosis and audit. Prompts and completions exchanged with AI model providers are written to the application log by default, and this is configurable per deployment. In a VPC deployment, all logs remain inside your cloud account.
Stored data and logs are retained under Telmai's data management policy retention schedule.
In a VPC deployment, retention is documented by data category. See Data Handling and Retention.
Your Controls
You decide what Telmai can reach. Connections are scoped by the credentials you provide, and Telmai recommends least-privilege, read-only access to source systems. You can configure prompt and completion logging, request read-only mode for AI write operations, or decline AI features entirely while retaining the full data quality function.
For data deletion requests or residency questions, contact info@telm.ai.
Last updated