More agents. Less RAM.

Native policy runner for memory-constrained Macs. Agent Lowmem keeps agent-launched validation predictable on Apple Silicon—one managed heavy operation at a time. Written in Rust.

Follow Pleo2 on GitHub v0.1 · early development · macOS arm64
agent-lowmem — managed session
$ agent-lowmem doctor
host
macOS arm64
profile
Apple M2 · 8 GiB
repository
JavaScript / TypeScript
policy
one heavy operation
watch
denied
cleanup
owned process group
ready
focused validation recommended
$ agent-lowmem run test
  1. inspectpackage.json + lockfile
  2. policytest · one worker · no watch
  3. lockacquired
  4. launchpnpm test
  5. supervisetimeout 15m · signals enabled
  6. cleanupprocess group released
  7. result completed

How it works

One guarded path from agent request to cleanup.

A managed runner is a supervisor for heavy commands. It does not replace Vitest, Jest, Next.js, npm, or pnpm—it wraps them and controls their lifecycle.

withoutagent → npm test → process fan-out → memory pressure
withagent → lowmem → policy + lock + supervision → controlled run

Global serialization

One Agent Lowmem-managed heavy operation across repositories, so cooperating agents do not pile work onto the same Mac.

No watch mode

Recognized watch and background execution are rejected before they can become forgotten memory consumers.

Bounded execution

Every supported operation has a visible deadline and a warning before Agent Lowmem ends it.

Owned cleanup

Timeouts and interruptions target only the process group Agent Lowmem created—never unrelated work by name or guesswork.

GitHub-aware inspection

agent-lowmem github inspect --json uses the official GitHub CLI to inspect Actions workflows through a bounded, read-only API request.