Tune device capabilities
Capability detection (chapter 2) measured what your hardware can do. This chapter is about declaring how much of it you're willing to lend Zyra at any one time. The right caps keep your own work snappy while still earning meaningfully.
Where the settings live
Devices → [your device] → Capabilities & limits in the dashboard. Each setting is enforced both by the agent locally and by the coordinator when matching jobs.
The five knobs
1. CPU cap
A slider from 10% to 100% of detected cores. Sets the maximum cores any task can use simultaneously. Maps to a Docker --cpus flag on each sandbox.
- 100% on a spare machine: maximizes earnings
- 50% on a daily-driver desktop: roughly invisible to your own work
- 25% on a laptop: stays cool and quiet
2. RAM cap
How much RAM tasks can use. Default 50%. Leave at least 4 GB for your OS and apps.
3. Disk allocation
Size of the working directory where the sandbox writes inputs, outputs, temp files. Default min(50 GB, 20% of free disk). Larger = qualifies for larger jobs.
4. GPU opt-in
Single toggle. Off by default. When on: qualifies for GPU-tagged jobs (ML inference, training, rendering) at materially higher rates [VERIFY: GPU rate multiplier at GA]. GPU runs near-100% — expect heat and fan noise. If you have a high-end GPU (RTX 4080+, Apple silicon 16+ GB), opting in is the single biggest earnings lever.
5. Max concurrent tasks
Default 1. Heavy machines (32+ cores, 128 GB RAM) can run 2-4 in parallel. Each gets a slice of the CPU/RAM caps. Leave at 1 unless you have headroom.
How caps affect capability score
The score reflects what you've made available, not your raw hardware. Cap a 16-core CPU at 25% and the coordinator sees 4 effective cores. Verified formula:
- Desktop/laptop: up to 300 from CPU + 300 from RAM + 400 from GPU = 1000 max
- Mobile/Android: up to 400 from CPU + 600 from RAM = 1000 max
Recommended starting profiles
| Profile | CPU | RAM | Disk | GPU | Concurrent |
|---|---|---|---|---|---|
| Spare machine, no daily use | 100% | 80% | 100 GB | On | 1-2 |
| Daily driver desktop | 50% | 50% | 50 GB | Off | 1 |
| Daily driver laptop | 25% | 33% | 25 GB | Off | 1 |
| High-end workstation | 75% | 60% | 200 GB | On | 2 |
| Apple silicon Mac mini | 75% | 50% | 100 GB | On | 1 |
Changes apply to the next task — running tasks finish under their existing limits.
Thermal and power considerations
Higher caps + GPU on = more heat. Monitor temperatures the first week. The agent exposes CPU and GPU temps on the device detail page. Stage 4 chapter 4 covers thermal limits in detail.
Electricity cost reality check
A modern CPU+GPU at full load draws 300-500W. At US average residential rate (~$0.16/kWh): 8 hours/day at 400W = ~$0.51/day = ~$15/month. The dashboard's Net earnings widget subtracts an estimated electricity cost based on a power-draw measurement during your first few jobs [VERIFY: power-draw measurement feature at GA].
Save and verify
Click Save. The agent receives the new caps within ~30 seconds. The dashboard "Effective capability score" updates immediately so you can see how your changes shift job-matching potential.
What's next
That closes Stage 2 — installed, paired, earning, scheduled, tuned. Stage 3 covers day-to-day operation: reading the dashboard, pausing, withdrawing, troubleshooting.
Last reviewed: 2026-05-21