AI agent
The console ships an AI agent (Dashboard → Onboard Your Agent hands out a prompt for *your* assistant; the ✨ button opens the built-in one). Both act through the same REST API with your identity, so nothing the agent can do is something you could not do yourself — and every call is checked by the same role and project rules.
Tools
Each tool is one API call. The agent's role decides which are offered: members see the read tools, admins and owners also get mutations, and destructive tools require approval.
Projects, environments & access — listProjects, getProject, listProjectServices, listProjectEnvironments, listProjectEnvironmentVariables, setEnvironmentVariable, deleteEnvironmentVariable, createEnvironment, updateEnvironment, deleteEnvironment, redeployProject, getProjectAccess, listProjectMembers, updateProjectMember, removeProjectMember, getProjectTransferPlan, transferProject.
Deployments — deployGitHubRepo, deployStarterApp.
Applications — listApps, getAppStatus, getAppLogs, getAppMetrics, scaleApp, rollbackApp, restartApp, stopApp, startApp, deleteApp.
Databases — listDatabases, provisionDatabase, provisionMongo, provisionRedis, provisionMySQL, provisionVectorDB, resizeDatabase, deleteDatabase, getDatabaseMetrics.
Object storage — listBuckets, createBucket, resizeBucket, presignUploadUrl, presignDownloadUrl, deleteBucket, getStorageMetrics.
Sandboxes — listSandboxes, getSandbox, createSandbox, updateSandbox, deleteSandbox.
Edge functions — listEdgeFunctions, deployEdgeFunction, createFunctionApiKey, listFunctionApiKeys, revokeFunctionApiKey, listFunctionVersions, getFunctionVersion, rollbackFunctionVersion, deleteEdgeFunction.
Networking & domains — createVPCIsolation, createGatewayRoute, configureFirewall, addCustomDomain, listCustomDomains, verifyCustomDomain.
Organization, keys & service accounts — listOrgInvites, createOrgInvite, listApiKeys, createApiKey, deleteApiKey, listServiceAccounts, createServiceAccount, createServiceAccountKey, enableServiceAccount, disableServiceAccount, deleteServiceAccount.
Audit, metrics & usage — listAuditEvents, exportAuditLog, getProjectMetrics, getUsage, getUsageLedger, getUsageActivity.
Approval workflow
Every mutating tool is scoped to admin or owner, and the destructive ones — deleteApp, deleteBucket, deleteDatabase, deleteEdgeFunction, deleteEnvironment, deleteEnvironmentVariable, deleteSandbox, deleteServiceAccount, transferProject — are marked high-risk and pause for explicit approval before they run. An approval is per-request, resolved by the signed-in user, and recorded alongside the action it authorised in the audit log.
Trust boundaries: the agent never sees a secret value (connection strings and credential blobs are stripped from tool output), it forwards your session or API key to the API rather than holding its own, and least-privilege filtering happens *before* the model can invoke a tool as well as again at the HTTP boundary.