Watch memory¶

One-off overview¶

clawtop

(clawtop status is the same.) The panel has three parts:

Bars. VM is the whole WSL/Linux machine. Gateway is the gateway's cgroup measured against its MemoryHigh limit, or against the VM when no limit is set.
Notes. Swap used and memory pressure; the gateway main process, the run stacks and other helpers; reclaimable page cache and the hard MemoryMax limit.
Run table. One row per agent run; see Concepts for what each column means.

Why the gateway bar is higher than the process sizes add up to

The cgroup counts page cache (files the kernel keeps in RAM, such as the SQLite databases). The notes line shows it as reclaimable. The kernel drops cache first when memory gets tight.

Live view¶

clawtop top

An interactive screen with sorting, filtering, marking, kill and a session tree. See Interactive view.

For scripts¶

clawtop status --json

prints one JSON object: memory figures, the cgroup limits, and a stacks array with each run's PID, agent, memory, age, idle time and member PIDs.

When to act¶

You see	Meaning	Do
Gateway bar near 100% of `high`	The kernel starts throttling and moving idle memory to swap	Stop runaway runs
`STALE` rows	Runs idle for 30+ minutes still hold 0.5 GB each	`clawtop kill --idle 30`
Swap climbing, pressure above a few percent	The gateway is running out of real RAM	Free memory now, then harden the gateway