Loops

Loops schedule recurring agent work without opening a terminal node and without interpolating shell commands. They run only while the BLXAde process is alive.

Open the Loops node

• Canvas Apps → Loops (create-or-focus; one node per project)
• Closing the node hides the UI; native schedules keep running until stopped or the app exits

Project vs global loops

Project catalogs use an opaque workspace id (not a raw absolute path). Moved projects may need an explicit rebind if the private index cannot reconcile them.

Scheduling model

• Fixed-rate future boundaries (no catch-up backlog after downtime)
• Start / Resume / Restart / interval edits arm at now + interval
• If a due run is busy or concurrency permits are full, the occurrence is skipped and the next future boundary is used
• Pause / Stop suppress future ticks while an active run finishes
• Restart, Remove, shutdown, and Rebind cancel the process tree

Concurrency is limited app-wide (worker + optional reviewer phases, with timeouts). Profiles are snapshotted at run start — agent setting changes apply on the next run.

Worker and reviewer adapters

Workers use verified direct one-shot modes only (examples: Claude -p, Codex exec, Cursor -p, Gemini -p, Grok -p, OpenCode run, Kimi -p where eligible). Unsupported wrappers are rejected.

Optional reviewers use narrower plan/read-only modes. Review prompts receive a condition plus a bounded masked worker tail — not full secrets.

Logs and safety

• Run logs are masked and stored with the loop store, not in the canvas drawing
• Disabled or missing agent profiles produce stable errors until fixed
• At startup, loop restore and profile sync can show a Prepare loops task in the loading UI before scheduling starts

When to use Loops

Open the node’s own controls for Start, Pause, Stop, Restart, interval, and definition edits — those actions are not duplicated on the node header menu.