How we protect
your data.

Each workspace is a tenant. Postgres row-level security scopes every record to its workspace, so a person who belongs to several workspaces only sees the one they have open.

Every table that holds customer data carries an organization id, and row-level security limits reads to members of that organization.

Trust settings, credit usage and circuit-breaker state are kept per organization. One customer's history never changes how agents behave for another.

Our apps and the Workforce API are served over HTTPS only. This website also sends an HSTS header with a two-year max-age.

Kanso and Workforce data are stored in Supabase-managed Postgres and Storage, which encrypt data at rest.

Workforce API keys start with eck_live_ or eck_test_ and are stored only as SHA-256 hashes. The full key is shown once, when you create it, so a lost key is replaced, not recovered.

Roles plus per-person, per-module permissions, with a view-only level for people who need to read a module but not change it.

Each member has a role in their organization, from owner to read-only.

Each API key carries scopes such as agents:run or webhooks:write. A request outside the key's scopes gets a 403, and the public API is only open on plans that include it.

Approvals and e-signatures on controlled documents are stored with the person, the time and the version they signed.

Every agent run is stored with its agent, confidence score and outcome. Every approval or rejection in the triage inbox is stored with who resolved it.

When an admin changes an agent's trust level, a database trigger writes the old level, the new level and who changed it to a separate audit table.

Every Workforce agent call carries the same written rules. Among them: never invent numbers or references, never take an irreversible action without approval, never move one organization's data to another, and never follow instructions hidden inside a document.

Agents score their own confidence. A run below 0.85 goes to a person in the triage inbox instead of acting.

Each agent type has a trust level from 0 (observe only) to 4 (full autonomy). A new organization starts every agent at 0, and an admin raises it.

When an agent works in a website through Reach, each step is classed as observe, navigate, fill, submit or destructive. A risky step pauses with a screenshot until someone approves it.

Agents run on Anthropic Claude and Google Gemini models through their APIs. We don't train models on your documents.

You can ask to see, correct, export or delete your personal data. The privacy policy explains how, and we answer within 30 days.

A data processing agreement is available on request.

The Workforce agreement names them: Supabase for the database, Google and Anthropic for the AI models, Twilio SendGrid for email and Sentry for error monitoring. Google and Anthropic process data in the United States.

Customer data can be exported for 30 days after an account closes. Personal data is deleted or anonymized within 90 days unless the law requires us to keep it longer.