@11agents/cli 0.1.32 → 0.1.33

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@11agents/cli",
-  "version": "0.1.32",
+  "version": "0.1.33",
   "description": "11agents local runtime and telemetry CLI",
   "type": "module",
   "bin": {

package/src/commands/runtime.js

@@ -83,6 +83,7 @@ function runtimeDeps(overrides = {}) {
     sleep: overrides.sleep || sleep,
     syncKnowledge: overrides.syncKnowledge || syncKnowledge,
     mcpKnowledgeSync: overrides.mcpKnowledgeSync || mcpKnowledgeSync,
+    cpuCount: overrides.cpuCount ?? os.cpus().length,
   }
 }
 
@@ -1540,9 +1541,10 @@ async function claimAndRunRuntimeTasks(registration, flags, deps, handlerModule,
   const config = configFromFlags(flags)
   const machineKey = registration?.machine?.machine_key || machineOverride(flags) || ''
 
-  // Each registered runtime must get at least one slot so no runtime is starved.
-  // maxConcurrent controls the ceiling but never drops below runtimes.length.
-  const effectiveConcurrent = Math.max(maxConcurrent, runtimes.length)
+  // Each runtime gets (vCPUs - 1) slots so the host stays responsive.
+  // maxConcurrent is honoured when it exceeds the CPU-derived default.
+  const perRuntime = Math.max(1, (deps.cpuCount ?? os.cpus().length) - 1)
+  const effectiveConcurrent = Math.max(maxConcurrent, runtimes.length * perRuntime)
   const slots = Array.from({ length: effectiveConcurrent }, (_, i) => runtimes[i % runtimes.length])
   const results = await Promise.allSettled(
     slots.map(runtime => runOneRuntimeTaskSlot(runtime, config, machineKey, registration, flags, deps, handlerModule, retryState, heartbeatIntervalMs))
@@ -1565,7 +1567,11 @@ export async function startRuntimeDaemon(flags = {}, deps = {}) {
   const scanIntervalMs = Number(flag(flags, 'scan-interval', '60')) * 1000
   const taskIntervalMs = Number(flag(flags, 'task-interval', flag(flags, 'heartbeat-interval', '15'))) * 1000
   const projectRefreshIntervalMs = Number(flag(flags, 'project-refresh-interval', '1800')) * 1000
-  const maxConcurrent = Math.max(1, Number(flag(flags, 'concurrency', '1')) || 1)
+  const cpuCount = resolvedDeps.cpuCount ?? os.cpus().length
+  const cpuDefault = Math.max(1, cpuCount - 1)
+  const envMax = Number(process.env.DAEMON_MAX_CONCURRENCY)
+  const envConcurrency = Number.isInteger(envMax) && envMax > 0 && envMax <= cpuCount - 1 ? envMax : null
+  const maxConcurrent = envConcurrency ?? Math.max(1, Number(flag(flags, 'concurrency', String(cpuDefault))) || cpuDefault)
   const once = Boolean(flags.once)
   const handlerPath = flag(flags, 'handler')