Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
Merge remote-tracking branch 'origin/dev' into feat/plugin-sdk-platform-rate-limits
Alfredo Lopez70edbedf4dfb
diff --cc control-api/src/config.tsindex a4dc68db2,b3b16ad83..6c588ac31--- a/control-api/src/config.ts+++ b/control-api/src/config.ts@@@ -130,16 -131,27 +131,35 @@@ type Config =approvalRlRequestPerMin: numberapprovalRlRefreshPerMin: numberapprovalRlExternalPerMin: number+ approvalRlExternalEdgePerMin: number+ approvalRlExternalClientIpPerMin: numberoauthBrokerRlPerMin: numberadminPublicTokenRlPerMin: number+ // Plugin Workload SDK platform rate limits (issue #348): per-minute+ // ceilings on the plugin abuse surface (data-path + pre-auth). Platform+ // protection, not per-service business quotas — usage/cost stays governed+ // by Token Budgets.+ pluginSdkNotificationsRlPerMin: number+ pluginSdkPromptBridgeRlPerMin: number+ pluginSdkRequestBucketRlPerMin: number+ pluginSdkPreauthRlPerMin: number+ // External Desktop GFS is a separate authority plane. The process-local+ // ingress backstop is intentionally wider than the distributed operation+ // buckets so normal Desktop read waterfalls cannot consume the security+ // budgets before authority resolution. The narrower values remain the+ // authoritative per-user/IP/actor operation ceilings.+ externalGfsIngressRlPerMin: number+ externalGfsTokenUserRlPerMin: number+ externalGfsTokenIpRlPerMin: number+ // Aggregate source-IP ceiling before authority resolution. This is a+ // defense-in-depth ceiling across authenticated non-token GFS traffic; it+ // is intentionally wider than the per-class/session/actor budgets.+ externalGfsIpRlPerMin: number+ // Read waterfalls have a separate budget from mutations. A Desktop root+ // refresh legitimately performs more reads than a mutation burst, while+ // mutation/grant/share ceilings remain deliberately narrow.+ externalGfsReadRlPerMin: number+ externalGfsOperationRlPerMin: number// Stateless-agent wake endpoint: per-host wake rate limit + server-side// coalescence window for the wake-annotation projection.hostWakeRlPerMin: number@@@ -720,26 -787,26 +795,43 @@@ export const config: Config =),approvalRlRequestPerMin: Number(process.env.APPROVAL_RL_REQUEST_PER_MIN || 120),approvalRlRefreshPerMin: Number(process.env.APPROVAL_RL_REFRESH_PER_MIN || 20),- approvalRlExternalPerMin: Number(process.env.APPROVAL_RL_EXTERNAL_PER_MIN || 60),+ approvalRlExternalPerMin: externalRateLimitConfig.operation,+ approvalRlExternalEdgePerMin: externalRateLimitConfig.session,+ // By default the source-IP backstop is wider than the per-session edge bucket+ // so a shared NAT does not turn the 120/min fairness budget into a+ // platform-wide ceiling. Operators may override these independent scopes;+ // crossed values are preserved after the boot advisory above.+ approvalRlExternalClientIpPerMin: externalRateLimitConfig.clientIp,oauthBrokerRlPerMin: Number(process.env.CONTROL_API_OAUTH_BROKER_RL_PER_MIN || 60),adminPublicTokenRlPerMin: Number(process.env.CONTROL_API_ADMIN_PUBLIC_TOKEN_RL_PER_MIN || 20),+ // Plugin Workload SDK platform rate limits (issue #348). Parsed via+ // positiveIntegerFromEnv so invalid operator config (non-integer, zero,+ // negative) fails loud at boot instead of running with NaN limits, which+ // would effectively fail open.+ pluginSdkNotificationsRlPerMin: positiveIntegerFromEnv(+ 'CONTROL_API_PLUGIN_SDK_NOTIFICATIONS_PER_MIN',+ 150+ ),+ pluginSdkPromptBridgeRlPerMin: positiveIntegerFromEnv(+ 'CONTROL_API_PLUGIN_SDK_PROMPTBRIDGE_PER_MIN',+ 120+ ),+ pluginSdkRequestBucketRlPerMin: positiveIntegerFromEnv(+ 'CONTROL_API_PLUGIN_SDK_REQUEST_BUCKET_PER_MIN',+ 600+ ),+ pluginSdkPreauthRlPerMin: positiveIntegerFromEnv('CONTROL_API_PLUGIN_SDK_PREAUTH_PER_MIN', 600),+ // Approved GFS authority-boundary budgets. Keep them fixed here rather than+ // accepting an unreviewed environment override. The 1800/min ingress guard+ // is only a coarse process-local backstop; the aggregate source-IP ceiling,+ // read buckets, token buckets, and mutation/delegation buckets remain+ // independently bounded below.+ externalGfsIngressRlPerMin: 1_800,+ externalGfsTokenUserRlPerMin: 10,+ externalGfsTokenIpRlPerMin: 600,+ externalGfsIpRlPerMin: 1_200,+ externalGfsReadRlPerMin: 120,+ externalGfsOperationRlPerMin: 30,// Default derived from the wake mechanism's worst case, not picked ad hoc.// rpc-proxy's wake-and-hold loop re-triggers POST /rpc/hosts/:hostRef/wake// every wakeRetriggerMs=15000 for up to wakeMaxHoldMs=90000 (defaults indiff --cc control-ui/package-lock.jsonindex 565e2ad6d,565e2ad6d..e57ac4765--- a/control-ui/package-lock.json+++ b/control-ui/package-lock.json@@@ -1,12 -1,12 +1,12 @@@{"name": "clerum-control-ui",-- "version": "0.1.425",++ "version": "0.1.426","lockfileVersion": 3,"requires": true,"packages": {"": {"name": "clerum-control-ui",-- "version": "0.1.425",++ "version": "0.1.426","dependencies": {"@clerum/display-field": "file:../packages/display-field","@clerum/llm-providers": "file:../packages/llm-providers",diff --cc control-ui/package.jsonindex 7093f5431,7093f5431..bca4d76e2--- a/control-ui/package.json+++ b/control-ui/package.json@@@ -1,6 -1,6 +1,6 @@@{"name": "clerum-control-ui",-- "version": "0.1.425",++ "version": "0.1.426","private": true,"scripts": {"dev": "NEXT_IGNORE_INCORRECT_LOCKFILE=1 next dev",