The shape of the system
Daemon Mode turns R1 into the harness and the loop at the same time. Instead of depending on external cron plus an orchestration wrapper, one long-running R1 process owns task admission, queue persistence, worker scheduling, WAL emission, underdelivery checks, and operator control. The failure boundary moves from shell glue into a single process with explicit state on disk.
Queued work moves into the pool, workers drain or underdeliver, and every outcome lands in the WAL before the loop restores the next task set. That is the claim: the same harness guarantees survive across hours, retries, and restarts.
Operator surface
| Surface | What it does |
|---|---|
POST /enqueue | Accepts new work without bouncing the daemon. Tasks land in the persisted queue immediately. |
GET /status | Reports worker counts, queue depth, pause state, and hook inventory. |
POST /workers | Raises or lowers concurrency at runtime. |
POST /pause / POST /resume | Stops admission into the pool, then restores the previous pool size later. |
GET /wal | Streams the append-only execution ledger used for audit and recovery. |
POST /hooks/install | Installs new safe hooks while the process stays hot. |
Why it matters
The claim is not “background jobs.” The claim is that the same harness-driven guarantees survive when work stretches across hours, retries, and restarts. Receipts still exist. Gates still exist. Underdelivery checks still exist. The loop no longer depends on external shell orchestration to remember where it was.