Performance
More headroom for AI work; intervenes later when pressure climbs.
by Humaan · macOS 14+
Linear
Keep memory pressure moving toward normal while you use AI tools.
v1.0.3 · menu bar · Apple Silicon (current build)
What it is
A governor for AI load — not a cleaner, not Activity Monitor.
A native macOS menu-bar governor that watches AI workloads (Cursor, Claude, Ollama, MCP, and related helpers), measures pressure, and relieves it safely — without casually killing your IDE or deleting model files.
Sole purpose: keep driving system memory pressure back toward normal during AI usage so the laptop stays usable. It may not always reach normal under heavy load, but every cycle works toward that.
Pipeline
How it works
01 · Discover
Enumerates processes and builds AI workload trees — IDEs, local models, MCP servers, language servers, and related workers.
02 · Measure
Tracks AI memory/CPU plus system pressure, swap, thermal, and battery. Deterministic heuristics only — no LLM in the loop.
03 · Detect
Flags runaway growth, idle-but-growing helpers, and high-confidence orphans — including helpers that are large versus this Mac’s usual size for that family.
04 · Relieve
Unloads idle models, clears orphans, and restarts idle Chrome, Cursor, and Docker helpers once an app passes 8 GB. Windows, GPU helpers, and the Docker VM stay up.
Protection modes
How aggressive relief gets
Balanced
Default posture — keep the laptop usable without starving tools.
Protect Mac
Earlier relief when pressure rises so the system stays closer to normal.
Behaviour
What you can rely on
Continuous pressure relief
When enabled (default), Linear works toward normal memory pressure during AI sessions — unload idle models, terminate confident orphans, renice large idle helpers.
Learns what worked on this Mac
After interventions, Linear measures reclaim and prefers tactics that worked here before. Failed tactics are demoted; learning never escalates. The banner can show a track record after a few samples.
Fail-open
Monitoring errors never kill user processes. Model files are never auto-deleted. Storage cleanup is always user-confirmed.
Local only
No accounts, analytics, or crash SaaS. Process info stays on the Mac. Core monitoring needs no network.
Requirements
Before you install
- macOS
- 14.0+
- Architecture
- Apple Silicon (shipped build)
- Sandbox
- Not App Store — needs process signals
- Signing
- Ad-hoc for now — Gatekeeper bypass required
Unsigned build — Gatekeeper bypass
Prefer the Terminal installer (clears quarantine). Browser downloads may need right-click → Open, or Privacy & Security → Open Anyway.
curl -fsSL https://dna.humaan.app/apps/linear/install.sh | bash
FAQ
Straight answers
Is this Activity Monitor or a RAM cleaner?
No. Linear is scoped to AI development workloads and system pressure. It is not a generic cleaner and not Electron.
Will it kill Cursor or Claude?
No. IDE processes are never auto-terminated. Interventions target helpers, orphans, and idle local models under policy.
Why does Gatekeeper block it?
Current builds are ad-hoc signed, not notarised. Prefer: curl -fsSL https://dna.humaan.app/apps/linear/install.sh | bash — or right-click → Open / Privacy & Security → Open Anyway.
Does Linear get smarter over time?
Yes, locally. It remembers whether a tactic actually reclaimed memory on this Mac and prefers ones that worked. It never uses that memory to take a stronger action than the current heuristics allow, and it never auto-kills the IDE.
How does this relate to DNA?
DNA is the engineering kit and Lab. Linear is a separate Humaan Mac app that keeps the machine usable while you run AI tools that DNA workflows often stress.
Related: Skeletor manages DNA Labs. Linear keeps the Mac usable while those tools run.
Get Linear
Menu bar AI resource governor for macOS. Prefer Terminal for a clean install on unsigned builds.
Prefer a clean install (skips the macOS “damaged” dialog for unsigned builds):
curl -fsSL https://dna.humaan.app/apps/linear/install.sh | bash