Agentlock: AI Agent Security

AI agents send emails, call APIs, modify databases, and trigger payments on your behalf. Without AgentLock, they do it with direct access to your credentials — and without asking.

AgentLock is a security-first approval platform that sits between your AI agents and the real world. Every action goes through you. Every credential stays encrypted. Every mistake can be undone.

REAL-TIME APPROVALS When an agent wants to act, you get a push notification instantly. You see the action, the target, and the risk level — categorized as Read, Write, Financial, or Admin. Approve with Face ID or deny with a tap. Hold-to-Approve prevents accidental confirmations. Unanswered requests expire automatically.

TIMELINE & UNDO Every agent action is logged in a detailed timeline — agent, action, timestamp, status, result. Like a bank statement for everything your AI does. When something goes wrong, hit Undo to reverse supported actions. Undo requires re-approval for safety.

ENCRYPTED CREDENTIAL VAULT API keys, passwords, and tokens are stored with envelope encryption. Your agents never see them — not in transit, not at rest, not ever. The secure runner decrypts only at the moment of execution, in memory. No API endpoint returns a decrypted secret.

POLICY ENGINE Define what gets approved automatically, what needs your confirmation, and what gets blocked entirely. Read actions pass through. Write actions require approval. Admin actions are blocked by default. Add domain allowlists for HTTP access.

BUILT ON SECURITY

• Ed25519 signed requests with replay protection
• End-to-end encryption with per-workspace content keys
• Row-level security on all database tables
• Append-only audit log — no deleting, no editing
• Session tokens stored in the iOS Keychain
• No tracking SDKs, no ads

WORKS WITH ANY AGENT AgentLock is agent-agnostic. Integrate using open SDKs for Node.js and Python. Manage agents, policies, credentials, and team members from the web dashboard at agentlock.net.