Security

Last updated:

This page summarizes how Mistle Cloud is currently designed to protect customer data. It is informational and may change as the service evolves.

For security questions or responsible disclosure reports, contact security@mistle.dev.

Service boundary

Mistle Cloud is the hosted Mistle service operated by Mistle. Self-hosted Mistle deployments are operated by their deployers and are outside the Mistle Cloud security boundary.

Mistle separates Mistle Cloud Providers from Customer-Configured Services:

  • Mistle Cloud Providers are vendors selected by Mistle to operate hosted infrastructure, analytics, billing, email, observability, secrets, and managed sandbox services.
  • Customer-Configured Services are third-party services a customer connects or configures for their organization, such as source-code hosts, chat systems, issue trackers, observability tools, or model providers.

Credentialless sandboxes

Mistle Cloud is built around isolated agent execution. Sandboxes are credentialless by default: supported integration credentials are not placed directly into sandbox environment variables or files.

Instead, managed outbound requests are mediated through the data-plane gateway. The gateway applies route policy, resolves the appropriate credentials through control-plane services, and injects credentials at request time. This helps agents use connected systems without directly seeing long-lived secrets.

This model does not stop customers from putting secrets directly into a sandbox themselves. Customers should avoid placing long-lived secrets in sandboxes unless they have reviewed and accepted the risk.

Integration secrets

Mistle Cloud encrypts integration secrets with organization-scoped keys protected by deployment master encryption keys. Runtime secrets for hosted infrastructure are distributed through managed secret systems rather than committed to source code.

Current Mistle Cloud Provider summary

CategoryProvider or systemTypical data handled
Edge, DNS, and static hostingCloudflareRequest metadata, edge routing, static app delivery
Runtime computeGoogle CloudHosted service traffic, workload runtime, operational metadata
Database and poolingPlanetScaleApplication records, organization data, session and integration metadata
Object storageGoogle CloudUploaded assets and sandbox storage objects where configured
EmailResendEmail addresses, transactional email content, delivery metadata
Product analyticsPostHogProduct usage events, technical identifiers, account or organization metadata
BillingStripeBilling customers, invoices, payment and tax details
ObservabilitySigNozLogs, metrics, traces, diagnostic metadata
Secret managementInfisicalRuntime secrets and deployment credentials
Sandbox executionE2B,
Tensorlake
Sandbox runtime metadata and execution infrastructure

Customer-Configured Services may also process Customer Content when customers connect them for their own workflows. Those services are governed by the customer’s configuration and by the service provider’s own terms and policies.

AI and model providers

Mistle does not use Customer Content to train Mistle or third-party foundation models.

Customer Content may be processed by AI, model, or agent runtime providers when a customer configures those providers or when processing is needed to run the requested agent workflow.

Monitoring and access

Mistle uses logging, metrics, traces, product analytics, and operational alerts to monitor reliability, performance, security, and abuse. Access to production systems and Customer Content is limited to what is needed for service operation, support, debugging, security, legal compliance, and incident response.

Customer responsibilities

Customers are responsible for:

  • selecting and configuring Customer-Configured Services;
  • choosing scopes, permissions, API keys, and connected accounts;
  • deciding what repositories, prompts, files, logs, and other Customer Content to submit;
  • reviewing agent output and actions before relying on them; and
  • maintaining backups, source control, and security controls appropriate for their use.