What we store

Agent data

Name, brokerage, market, price band, subdivisions, contact details and bio — the fields an agent enters to provision a publication. Most of this is public by design because it appears on the publication. Email and password authentication is handled by Supabase; passwords are never stored in this application.

Content and tokens

Posts, closed transactions and uploaded images are stored so they can be published. MCP access tokens are stored only as SHA-256 digests — the plaintext is shown once at issuance and cannot be recovered. Rotating a token revokes the previous one immediately.

Processors

• Supabase — authentication, Postgres and object storage.
• Google Gemini — receives the structured inputs for a draft (market parameters, subdivision names, the agent's notes) to generate post text. It does not receive account credentials.
• Vercel — application hosting where deployed.

Cookies and tracking

One session cookie for the dashboard, and one for sidebar state. No advertising trackers, no third-party analytics scripts, no cross-site profiling. The publication keeps its own cookie-less request counts — which page was asked for, the referring host and whether the client is a person or a crawler — so the agent can see what was read. No IP address and no full user agent is stored. Embed widgets set no cookies; their attribution links carry a utm parameter so referrals can be counted server-side.

Retention

Content stays until the agent removes it or the account is closed. MCP call logs are kept to a rolling window and hold the tool name, outcome and duration — not the arguments. Search-engine crawlers may retain copies of public pages independently of this application.