Security & Data Handling

Your business memory is portable

Chief's Command holds a structured record of the decisions, promises, standing instructions, facts, and open loops that matter to your business. Supported AI tools can access the approved business memory through a secure connection. If you switch AI models tomorrow, your approved context comes with you.

What connected AI can and cannot access

• Can access: the client workspace's active memory records, ventures, ledger, open loops, decisions, people you've entered, and any skill packs you've enabled.
• Cannot access: another client's data, revoked installations' state, raw chat transcripts of your other conversations, or your complete AI history on those platforms.
• Connected AI can only share data with Chief's Command through explicit tool actions it takes. Chief's Command does not read your other chats or your full AI history.
• No raw chat transcripts are stored by default — only the structured summaries the AI intentionally records.

Tenant isolation

• Every row is scoped to a client workspace. Postgres row-level security enforces this at the database layer, not just in application code.
• An AI installation belonging to one client cannot read, write, or even see the existence of another client's data.
• The public demo workspace is walled off from real workspaces at the policy layer.

Installation identity and presence

When you connect an AI, it registers itself and receives a one-time instance token. Only a SHA-256 hash of that token is stored server-side; the raw token is kept by your assistant in whatever mechanism it uses to persist per-workspace secrets. Revoking an installation invalidates the token immediately and releases any tasks it was holding. Exactly one installation per workspace can be marked primary; only the primary receives the daily brief.

Presence (online / idle / offline) is derived from installation heartbeats and is best-effort. Messages targeted at an installation that is currently inactive remain queued until it next checks in — Chief's Command does not push messages to sleeping installations.

Authentication

• Human sign-in: email + password, or Google OAuth.
• AI installations authenticate via OAuth 2.1 and act as the signed-in user; privileged operations require a verified session.
• There are no unauthenticated write paths into your workspace.

Auditability, export, and deletion

Every write to memory is captured in a per-client audit trail (who, when, before/after). You can review the trail from Chief's Command HQ, export the workspace's memory, correct or supersede records, and request full deletion.

Platform

Chief's Command runs on managed cloud infrastructure. The platform provides encryption in transit for network connections and managed Postgres for the database. Chief's Command does not claim any specific compliance certification, backup SLA, uptime SLA, or encryption-at-rest guarantee beyond what the underlying platform provides at any given time.

Limitations

• AI can produce errors, omissions, or unexpected output — connected assistants may occasionally record something incorrectly. Review the memory page and correct as needed.
• Third-party AI platforms (Anthropic, OpenAI, Google, etc.) operate under their own terms and privacy policies for the chats that occur on their side.
• Instance presence and message delivery are best-effort; queued messages are delivered when the target installation next checks in.
• No system is perfectly secure. Nothing here is a guarantee against every possible failure mode.

Contact

For privacy or security questions, contact josh@rosenfeldmarketing.com.