Use cases
Where a pre-execution guard earns its place
Agents that can change things: run shell commands, write to databases, move money, send email or touch infrastructure. Each page shows what can go wrong, the check to send, and the controls to keep alongside it.
Checks run on our GPUs in Switzerland. Request payloads are not stored.Security
6 pages · press / to search
- Use caseDeveloper agentsGuarding coding agents before they touch your shell, git and filesHow to check a coding agent’s shell, git and file operations before they run: rm -rf, force pushes, secret reads and outbound curl, with allow, ask or block.Read
- Use caseDataGuarding database agents before their SQL runsCheck an agent’s SQL before it runs: DROP and DELETE without WHERE, production versus staging, and bulk exports of personal data, with allow, ask or block.Read
- Use caseCustomer operationsGuarding customer-support agents that issue refundsCheck a support agent’s refund calls before they run: amounts not in the ticket, refunds to the wrong customer and instructions injected through ticket text.Read
- Use caseProductivityGuarding email and calendar assistantsCheck an email or calendar assistant’s actions before they run: external recipients, forwarded inbox contents, instructions hidden in email and deleted events.Read
- Use caseDeveloper agentsGuarding infrastructure and DevOps agentsCheck an infra agent’s actions before they run: terraform destroy, kubectl delete, scaling to zero, printing secrets and IAM changes, with allow, ask or block.Read
- Use caseSandboxed agentsGuarding code-mode agents at the sandbox tool proxyIn code mode, generated code fires many tool calls no human sees. Guard every call at the sandbox tool proxy, in batches where possible, before it runs.Read