System Requirements
Arble itself is a CPU workload. GPUs serve model inference and, optionally, embeddings — if your models are remote, you need no GPU at all.
| Minimum | Recommended | |
|---|---|---|
| CPU | 4 vCPU | 8 vCPU |
| RAM | 8 GB | 32 GB |
| Storage | 40 GB SSD | 250 GB NVMe |
| GPU | None | 1 × 24 GB, for local models |
| OS | Linux 5.15+, macOS 14+ | Linux 5.15+ |
Operating systems
Ubuntu 22.04 LTS and 24.04, Debian 12, RHEL 9 and derivatives are tested. macOS 14+ is supported for desktop and workstation use. Windows is supported through WSL2 for development, not production.
Kernel requirements
Tool sandboxing uses cgroups v2, user namespaces and seccomp. On a kernel older than 5.15, or one with user namespaces disabled, Arble falls back to weaker isolation and logs a warning at startup — it degrades rather than failing, which is worth knowing before you rely on the sandbox.
arble server verify-sandboxStorage characteristics
The database is latency-sensitive on writes; put it on NVMe or provisioned IOPS. Object storage is throughput-sensitive rather than latency-sensitive, so network storage is fine there.
Plan roughly 2 GB of database per 10,000 runs retained. Files and screenshots dominate total volume, not run records.