# Security and Trust for AI Agents > Where Sidekick agents run, how they sign in, which actions wait for a person, what goes on record and how to stop them. Security reviews on request. Source: https://sidekickmachines.com/trust/ Trust and security ## Your agents, in your cloud. Every step on record. Answers for security reviewers and operations leaders: where agents run, how they sign in, what waits for a person, what goes on record and how to stop them. [Send your questionnaire](mailto:security@sidekickmachines.com?subject=Security%20questionnaire) See who does what At a glance 1. Runs in your cloud 2. Signs in as itself 3. A person approves 4. Every step on record Where agents live ### Agents run where your data lives. License Sidekick Sovereign, or have our engineers build in your environment. Either way, your agents run in your cloud, and your people reach them in Microsoft Teams. No SaaS. Zero bits flow to us. You keep 100% of everything, keys included. [See agent operations](https://sidekickmachines.com/agent-operations/) Engineering in your environment #### Your own environment Our engineers work inside your environment. Your agents run in your cloud, next to your data. Your cloud - Your agents - Your data - Our engineers, inside Licensed Sidekick Gateway and Sidekick Memory run in your own environment too. Sidekick Sovereign #### Your own Microsoft Azure Every Sidekick product as your own private copy, shared with no other customer. AI models run through your own Microsoft Foundry. Your Microsoft Azure - Sidekick Gateway - Agent Studio - Sidekick Memory, add-on - Sidekick Connectors, add-on [Sidekick Sovereign](https://sidekickmachines.com/products/sovereign/) Plant networks: Sidekick Connectors move data outbound only, so your plant network never opens up to cloud services. [Sidekick Connectors](https://sidekickmachines.com/products/connectors/) Who does what ### Who does what, in plain words. Both setups, side by side. Your keys stay with you in both. | Way of working | Who hosts | Who operates | Who holds keys | Who can stop agents | | ---------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- | ----------------------------------------------------- | ------------------------------------------------ | --------------------------------------- | | [Sidekick Sovereign in your Microsoft Azure](https://sidekickmachines.com/products/sovereign/) | You, in your own Microsoft Azure. A private copy, shared with no other customer. | Us, as our core service. Or your team, if you prefer. | You. Always. Agents never see passwords or keys. | You. Any agent, or all of them at once. | | Engineering in your environment | You, in your own cloud. Our engineers work inside your environment. | Us, as our core service. Or your team, if you prefer. | You. Always. Agents never see passwords or keys. | You. Any agent, at any time. | How agents sign in ### Each agent signs in as itself. Never with a person’s password. Agents never see passwords or keys. - #### Its own identity Every agent has its own identity and never borrows a person’s, so your records show which agent acted. - #### Access by role Each agent can use just those tools and actions approved for it. Anything else is refused. - #### Access follows its job We scope its access to its job and revoke access when that job changes. - #### Older systems too Modern and legacy sign-in methods both work, so older enterprise systems can join in. With Agent Studio, each new agent gets its own Microsoft Entra identity and Microsoft 365 account. [Agent Studio](https://sidekickmachines.com/products/agent-studio/) What needs a person ### Sensitive actions wait for a person. A person approves that exact request. One approval covers one request, and nothing more. - #### You choose what waits You tell us which decisions stay with a person. We set which actions wait for approval. - #### One approval, one request An approval covers that one request. A different request waits for a person again. - #### Every decision on record Who approved what, and when. New agents need sign-off too: in Agent Studio, choose none, one or two approvers, as your policy says. Nobody approves their own request. Example: a finance agent asks to pay invoice 1042. It waits until an accounts payable lead approves that exact request, and that approval goes on record. When it then asks to pay invoice 1043, that new request waits for a person again. What goes on record ### Every step, on record. Every request, check and approval is recorded: who approved what, and when. Example audit trail: each line shows a time, an agent or person, a request and its result. A maintenance agent reads alarm history and raises a work order (permitted), asks to order a spare part over its limit (needs approval), a shift supervisor approves it, and a finance agent’s request outside its role is denied. - #### Ready for auditors Every action, handoff and approval is on record, ready when auditors ask. - #### Export it Agent Studio keeps a tamper-evident audit trail you can export. - #### Memory too Everything Sidekick Memory remembers and recalls stays on record. Stopping agents ### Stop any agent, or all at once. You can stop any agent at any time. With Agent Studio and Sidekick Sovereign, one click stops one agent, or every agent at once, with a full audit trail. - #### Every agent in one place See every agent in one place, and stop any one of them with one click. - #### Others keep working Sidekick Gateway has no central gateway, so a problem stays contained and other agents keep working. Example: three agents running. One click stops a finance agent while a maintenance agent and a planning agent keep working. Stop all stops every agent at once. For security teams ### More detail for your security review. Product #### [Sidekick Gateway](https://sidekickmachines.com/products/gateway/) A decentralized MCP gateway: human approval for sensitive actions, every request on record. See Sidekick Gateway Product #### [Sidekick Sovereign](https://sidekickmachines.com/products/sovereign/) Every Sidekick product as a private copy in your own Microsoft Azure. See Sidekick Sovereign Engineering #### [Agent Security](https://sidekickmachines.com/capabilities/agent-security/) Put identity, permissions, human approvals and secure execution into practice. See Agent Security Reviews and privacy ### Security reviews and privacy #### Security questionnaires and reviews Security questionnaires and reviews are available on request. Send yours, or ask for a call with our engineers. [Send a questionnaire](mailto:security@sidekickmachines.com?subject=Security%20questionnaire) #### Privacy This website sets no cookies and loads no third-party scripts. Customer engagements run under their own signed agreement, which covers data handling, security, residency and audit terms. [Read our privacy policy](https://sidekickmachines.com/privacy/) Responsible disclosure ### Found a security issue? Tell us. Think you’ve found a vulnerability in our website or software? Here’s how to report it. 1. #### Write to us Email security@sidekickmachines.com with subject “Security issue”. 2. #### Share details What you found, where, and steps to reproduce it. 3. #### Allow time for a fix Please give us time to fix an issue before you share details publicly. While testing, please stay clear of other people’s data and avoid disrupting service. [Report an issue](mailto:security@sidekickmachines.com?subject=Security%20issue) This contact also appears in our [security.txt file](https://sidekickmachines.com/.well-known/security.txt) Security review ### Bring your security questions. Send your questionnaire or ask for a review call. Our engineers answer in plain words. [Email your questions](mailto:security@sidekickmachines.com?subject=Security%20review) Microsoft, Microsoft 365, Microsoft Azure, Microsoft Entra, Microsoft Foundry, and Microsoft Teams are trademarks of the Microsoft group of companies.