Headroom

SECURITY

Guardrails at every layer.

Headroom is built for operational data and consequential decisions. Isolation, visibility, and human authority are part of the architecture.

Tenant isolation

Every customer-data record belongs to an organization. Each request opens a tenant-scoped database session, and PostgreSQL row-level security (RLS) enforces the organization boundary again. RLS is forced even if application code makes a mistake.

Encryption and identity

Data is encrypted in transit between your browser, Headroom, and our service providers. Users authenticate with Google or Microsoft; Headroom does not store their passwords. Organization roles limit who can view, change, approve, or administer work.

The Constitution

Every organization sets a maximum automatic spend. A purchase order above that limit is always pending approval. This rule is hard-coded: an agent, an LLM, and an administrator changing other settings cannot override it.

Glass Box accountability

Agent actions create structured, organization-scoped logs that show when an action happened, which agent acted, what it reported, and the event type. Users can watch, pause, and overrule governed workflows. Deterministic code owns math and decisions; LLMs only advise.

Customer data and AI

Headroom does not use customer data to train models. Authorized advisory requests may send only the relevant context to third-party LLM providers under business or API terms that prohibit training on submitted customer data.

Deletion and reporting

Organization owners may request data deletion. After verifying authority, we remove or de-identify customer data from active systems and let protected backups expire on their normal schedule, except where retention is legally required.

To report a security concern, contact [SECURITY CONTACT EMAIL]. Please include enough detail for us to reproduce the issue and avoid accessing data that is not yours.