How we protect your agency's and clients' data
Agency isolation, private files, encrypted secrets, and the guardrails on agents that change client code.
Agencies are isolated from each other
Every read and change is checked against your agency and the client's organization. Database rules, not just the screens, stop one agency's staff from seeing or changing another agency's data, and stop clients from seeing anything outside their own projects.
Internal notes on proofs and projects are hidden from clients by the database itself.
Files and downloads
Proofs, brand kit files, briefs, and review recordings are stored privately. Downloads are checked against project access and use links that expire within minutes. Your agency's logo and branding images are public, because they appear on your sign-in pages and in emails.
Secrets and connectors
Connector secrets, such as Slack webhook addresses and Google access, are encrypted with AES-256-GCM before they are stored, and no browser session can read them back. Site environment variables marked sensitive can never be read again.
Agents that change client code
- Agents never run inside the app. Each change runs in an isolated job with access to one client repository, using a token that expires within the hour.
- A leak guard rejects any change that touches workflow, agent, or environment files, or that adds anything shaped like a credential.
- Nothing an agent builds merges by itself. Every change needs your team's review and your client's approval before it is published.
Still stuck? Email hello@modelmerge.ai with your agency name, the client and page involved, and the exact message you see.