Brooke handles your business data — emails, documents, CRM records. We take that responsibility seriously. Here's exactly how we protect it.
Brooke runs entirely on Cloudflare's platform — Workers, D1, and R2. This is not a traditional cloud deployment. Every component runs at the edge, globally distributed, with enterprise-grade security at every layer.
We never train AI models on your data. Your runs, artifacts, and integration content are never used to improve our AI models or anyone else's. Full stop.
Every workspace is fully isolated at the database and storage layer. Your data is scoped to your workspace and cannot leak across customers — by architecture, not just policy.
Artifacts stored for 90 days by default. Run logs retained for 12 months. You can delete your workspace data at any time — deletion is immediate and permanent.
OAuth tokens and API keys are encrypted at rest and scoped to your workspace only. No cross-workspace key sharing.
Brooke requests only the minimum OAuth permissions needed for each integration. We never request broad access when narrow scopes suffice.
Admin, member, and viewer roles with granular permission controls. Admins control who can connect integrations, approve actions, and access billing.
Disconnect any integration, revoke any API key, or delete your entire workspace at any time from settings. Deletions are immediate and irreversible.
Brooke processes business content through AI. We enforce strict safeguards at every step — from what the model sees, to what it's allowed to do, to what it's allowed to say.
User data, retrieved content, and connector payloads are treated as untrusted. Strict boundary separation prevents override of system instructions — even if injected into documents or emails.
The model only sees data relevant to the current run. No cross-workspace context, no cross-user history. Each run starts with a clean, scoped context window.
All AI-proposed tool calls are validated against schema and policy before execution. Malformed, out-of-scope, or policy-violating tool calls are rejected before they can execute.
Before any outbound action, content is checked for patterns consistent with attempted data exfiltration. Suspicious outputs are flagged and blocked before delivery.
Every external action — sending email, updating CRM, publishing content, creating calendar events — requires explicit human approval before execution. This isn't a feature. It's a constraint.
You are always in control. Brooke never takes irreversible actions without your explicit consent. Approval requests include a full preview of the intended action — exactly what will be sent, modified, or created.
Formal audit in progress. Target completion: Q3 2026. Results will be shared with enterprise customers upon request.
Data processing agreements available for EU/EEA customers. Standard Contractual Clauses in place for international transfers.
California Consumer Privacy Act compliant. We do not sell personal data. California residents can exercise rights at privacy@brooke.com.
Custom Data Processing Agreements and Business Associate Agreements available for healthcare, finance, and enterprise customers.
We take security vulnerabilities seriously and appreciate responsible disclosure. If you've found a potential security issue in Brooke, we want to hear from you.
Our security team is happy to answer questions, provide documentation, or walk through our architecture in detail.