Skip to Content
Platform

Data handling

What we hold from your agent runs, how we store it, and what control you have.

What we hold

Introspection captures what your agents do when they run on the platform. The records you work with are built from the run data the runtime emits:

  • Tasks & conversations: the replayable record of a task. Model calls, tool calls, and messages.
  • Observations and patterns: platform-generated records about completed conversations, and clusters of similar observations across many runs.
  • State & artifacts: files runtimes read and write, the durable artifacts they produce, and read-only share links.
  • Underlying telemetry: the events, model/token metadata, and timing beneath those records, plus user identity and feedback signals attached to a run.

What we do not do

  • We do not train AI models on customer data. Introspection’s internal models are trained on separate data. Customer data is used only to operate the service.
  • We do not sell or share personal data with third parties for advertising.
  • We do not use advertising or tracking cookies. Only essential session cookies are used.

Data residency

Where your agent run data lives depends on your deployment model:

  • Managed Cloud: data is held in Introspection-operated cloud infrastructure.
  • BYOC: agent-run data is held in the cloud account and region you control.
  • On-prem: data remains inside the private or air-gapped environment defined by the Enterprise deployment.

See Deployment for the full picture.

Data retention

Retention periods are governed by your subscription plan. Contact support@introspection.dev for details on your plan’s retention policy.

Customer-managed encryption keys

BYOC and on-prem deployments can support customer-managed encryption keys through the deployment’s key-management system. When enabled, customer-managed keys cover stored data: databases, caches, object storage, and Kubernetes secrets. Availability depends on the target infrastructure; contact us to confirm support for your setup.

Sensitive data

The platform is designed for agent run data: model inputs and outputs, tool calls, files, and user feedback. Avoid routing sensitive PII (social security numbers, financial account numbers, health records) through agent runs unless you have an enterprise agreement in place.

If your application processes sensitive data and you need to redact it before it reaches Introspection, contact us about PII handling options.

Subprocessors

Introspection uses the following categories of subprocessors:

CategoryExamples
Cloud hostingAWS, GCP, Azure
Databases & infrastructurePostgreSQL, ClickHouse, Valkey, Restate (self-hosted)
AuthenticationZitadel (self-hosted)
Model providersAnthropic, OpenAI, Google Gemini
PaymentStripe

For a complete list, contact support@introspection.dev.

Last updated on