Zero-secret vault
Keep API keys out of the agent runtime. Agents request actions and AgentLock's secure backend runner performs the authorized work.
Agentlock
AgentLock is a security gateway for AI agents that protects API keys and other credentials with a Request-Approve-Execute workflow. It helps individuals and teams review sensitive agent actions, enforce policy, and keep an auditable execution trail.
Overview
AgentLock is a security gateway for AI agents. It sits between agents and sensitive systems so actions can be reviewed before anything is executed, rather than giving agents direct access to API keys or passwords.
The product is built around a Request-Approve-Execute workflow. Agents send cryptographically signed requests, humans approve them from mobile, and a secure runner executes the task while returning sanitized results. The result is a controlled execution layer with policy checks, audit history, and rollback for supported actions.
Core features
Mobile approval gateway
Review the exact action details on mobile and approve or deny requests in real time, including with biometric authentication and push alerts.
Smart policy engine
Apply workspace rules for read, write, and financial actions, including stricter approval requirements for higher-risk operations.
Immutable audit trail
Track every request and approval in an end-to-end encrypted timeline that acts like a reversible record of activity.
Multi-user approvals
Require multiple people to sign off on high-risk actions and assign custom roles for team workflows.
Undo and safety controls
Roll back supported actions from the dashboard and use the emergency kill switch or safe mode to stop or restrict ongoing agent activity.
Use cases
Protecting sensitive credentials
Use AgentLock when an agent needs access to a sensitive API key or password but should never see the raw secret. The gateway injects credentials only during authorized execution.
Reviewing high-risk agent actions
Use it for operations that should be reviewed before they execute, such as write or financial actions. Humans can approve or deny each request from mobile.
Team-controlled operations
Use it when multiple people need to approve a change before it goes live. Team roles and multi-user approvals support a four-eyes workflow for sensitive tasks.
Auditing and recovery
Use it when you need a record of what an agent requested and what was approved. The immutable timeline supports audits, troubleshooting, and rollback for supported actions.
Policy-based automation
Use it for constrained automation where safe tasks can run automatically but risky tasks should wait for review. The policy engine can allow read-only work and gate higher-risk actions.
Pros and Cons
Pros
- Keeps credentials hidden from agents by routing work through a protected gateway and secure runner.
- Supports real-time human approval from mobile, which is useful for time-sensitive review.
- Adds policy controls for read, write, and financial actions instead of treating every task the same.
- Includes an encrypted activity timeline and rollback for supported actions, which helps with review and recovery.
- Offers team-oriented controls such as multi-user approvals, custom roles, and a kill switch.
Cons
- The undo feature only applies to supported actions; the source explicitly notes that some actions, such as sent emails, are permanent.
- If a required approval is not granted within 5 minutes, the request is blocked, so unattended workflows can stall.
- The source does not document broader integrations, deployment options, or team setup details beyond the listed vault use cases and provider examples.
FAQ
How does the approval workflow work?
AgentLock routes agent actions through a Request-Approve-Execute flow. The agent sends a cryptographically signed request, a human reviews the action details, and the secure runner executes approved work.
What happens if I do not approve a request in time?
If a manual approval is required and not granted within 5 minutes, the request is automatically blocked.
Does my agent ever see my API keys?
No. The agent sends a request to the gateway, and AgentLock injects the credential in a protected environment during execution so the raw secret is not handled by the agent.
What kinds of credentials can I secure?
AgentLock says it can manage API keys, login passwords, email access, and other credentials for services an agent needs to interact with.
Can I undo an executed action?
Supported actions can be reverted from the dashboard, but the source notes that some actions, such as sent emails, are permanent.
Quick Facts
- Category
- AI agent security gateway
- Website
- agentlock.net
- Workflow
- Request-Approve-Execute
- Pricing
- Free, Pro, and Teams plans listed
- Plan notes
- Annual billing shows a 30% saving; custom plans are mentioned
- Platform notes
- Mobile approvals with secure runner execution
Alternatives à Agentlock
PromptScout
PromptScout tracks how ChatGPT, Gemini, Google AI Overviews, and Perplexity mention your brand or competitors, then pairs those results with source analysis and website audits. It helps teams decide what to fix in content, positioning, or site readiness next.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
Sleek Analytics
Sleek Analytics is a privacy-friendly web analytics tool with real-time visitor tracking, Core Web Vitals, and revenue attribution. It helps site owners understand traffic and conversions without cookie banners or a heavy setup.
GitBot
GitBot is a local browser-based workspace for creating reusable coding bots powered by Claude Code, Codex, or OpenCode. It helps developers run those agents across repository folders with persistent threads and configurable permissions.
Codex Plugins
Codex Plugins bundle reusable skills, app integrations, and MCP servers into workflows you can install in the Codex app or use from Codex CLI. They help extend Codex with connected-service tasks, reusable instructions, and shared team workflows.
MacSpoof
MacSpoof change ou randomise l’adresse MAC Wi‑Fi sur macOS pour reconnecter aux réseaux et limiter l’enregistrement de votre identité sur Wi‑Fi public.