diff --git a/apps/api-daemon/src/capture/streamIngest.ts b/apps/api-daemon/src/capture/streamIngest.ts
index 4db287e..3fcb75e 100644
--- a/apps/api-daemon/src/capture/streamIngest.ts
+++ b/apps/api-daemon/src/capture/streamIngest.ts
@@ -181,8 +181,12 @@ export function startStreamIngestWorker(): Worker<StreamIngestJob> {
     // count; the real ceiling is the server's CPU/bandwidth for running that
     // many yt-dlp+ffmpeg pairs at once, not this number.
     concurrency: env.maxConcurrentCaptures,
-    // Captures run for the lifetime of the stream (can be hours) — extend the
-    // default lock well beyond BullMQ's stalled-job assumptions for an MVP.
-    lockDuration: 24 * 60 * 60 * 1000,
+    // BullMQ renews this lock automatically while a worker is alive, so a
+    // low value doesn't cap how long a capture can run — it only controls
+    // how fast a *dead* worker's job is detected as stalled and recovered.
+    // (An earlier 24h value meant a killed process's job wouldn't be
+    // recovered for up to 24h; closeOrphanedSessions() in index.ts is what
+    // actually unblocks that channel's StreamSession in the meantime.)
+    lockDuration: 10 * 60 * 1000,
   });
 }
diff --git a/apps/api-daemon/src/index.ts b/apps/api-daemon/src/index.ts
index de28a1f..1570139 100644
--- a/apps/api-daemon/src/index.ts
+++ b/apps/api-daemon/src/index.ts
@@ -30,7 +30,28 @@ async function forceLiveBypass(youtubeUrl: string) {
   });
 }
 
+/**
+ * A container restart (e.g. a redeploy) kills any in-progress capture
+ * without running its cleanup code, leaving that StreamSession's endedAt
+ * stuck at null forever. Besides showing "still recording" in the panel,
+ * startSession() treats any open session as a reason to skip starting a new
+ * one for that channel — so an orphaned session permanently blocks that
+ * channel from ever being recorded again until this runs. Since a fresh
+ * process start means no capture can actually still be in progress, every
+ * open session at boot is stale by definition and gets closed out.
+ */
+async function closeOrphanedSessions() {
+  const { count } = await prisma.streamSession.updateMany({
+    where: { endedAt: null },
+    data: { endedAt: new Date() },
+  });
+  if (count > 0) {
+    console.log(`[index] closed ${count} orphaned session(s) left open by a previous process`);
+  }
+}
+
 async function main() {
+  await closeOrphanedSessions();
   startServer();
   startStreamIngestWorker();