Solutions

What you can run on a Deliany Mac.

Three things Apple Silicon is the best machine in the world for — and we rent them by the hour, the month, or the year, from a rack in East Austin.

AI & inference iOS & macOS CI/CD Emerging use cases Hardware

AI & inference

LLMs that fit in unified memory, billed by the hour.

For teams shipping LLM features that need private inference, low cost-per-token, or a real test bench for Apple Intelligence workflows. Our Max tier ships with MLX, MLX-LM, llama.cpp (Metal), Ollama, LM Studio headless, Core ML tools, and Python 3.12 via uv — all preinstalled and ready.

Common configurations

MLX Studio — MLX, MLX-LM, Jupyter on :8888, Llama 3.3 70B 4-bit cached, ready in 8 min.
Ollama Server — Ollama HTTP API on :11434, Llama 3.3 70B + Qwen 2.5 32B preloaded.
LoRA Trainer — MLX fine-tune tooling, HuggingFace cache on 1TB volume, WandB env preset.
RAG Stack — Ollama + Qdrant + a 48GB mmap-friendly volume for embeddings.

Honest limit: training large models from scratch is still an NVIDIA job. We're built for inference, LoRA fine-tunes, and anything that fits in 64–128GB of unified memory. For the rest, we'll happily point you at the right tool.

# provision an M4 Max from the CLI
$ deliany launch --tier max --image mlx-studio
→ m4max-austin-042  online in 6m 41s

$ ssh m4max-austin-042
➜ ollama run llama3.3:70b
>>> Write a haiku about unified memory.

  One pool of silicon,
  seventy billion parameters
  resting in the same wafer.

total duration:   43.2s   (~11 tok/s)
eval count:       476 tokens
load duration:    0.9s

# .github/workflows/ios.yml
name: iOS Build & Test
on: [push, pull_request]

jobs:
  build:
    runs-on: [self-hosted, deliany-m4pro]
    steps:
      - uses: actions/checkout@v4
      - name: Install deps
        run:  bundle install && pod install
      - name: Build + Test
        run:  bundle exec fastlane test
      - name: Archive & Notarize
        if:   github.ref == 'refs/heads/main'
        run:  bundle exec fastlane beta

# a clean Xcode 16 build on M4 Pro: ~6 min.
# an equivalent shared runner: 28 min.

iOS & macOS CI/CD

A build farm that actually feels like a Mac.

For iOS teams whose Xcode Cloud bill has a comma in it, or whose notarization pipeline keeps breaking on shared runners. Every Core and Max tier is a bare-metal Mini you can SSH into — one tenant, root access, your keychain, your profiles.

What ships preinstalled

Xcode 16.x, command-line tools, simulators, notarization kit
fastlane, Tuist, xcbeautify, cocoapods, swift-format
Homebrew, gh, GitHub Actions runner, Bitrise CLI
Per-branch ephemeral Minis via our CLI and HTTP API

Compliance

SOC 2 Type I on the roadmap for Q3 2026 — we can share preliminary controls documentation and sign an MSA today. Every Core and Max tier is physically isolated: one tenant per machine, full-disk reimage between tenants.

Emerging use cases

Other things our customers do with a Mac they can hold down for a month.

visionOS

Vision Pro hosts

M4 Pro with real Vision Pro simulator performance and Thunderbolt passthrough for tethered devices.

Agents

Agent sandboxes

Clean macOS snapshots for Claude and OpenAI computer-use agents. Snapshot-and-reset via API.

Creative

Remote stations

Final Cut, Logic, and DaVinci Resolve with GPU acceleration the way Apple intended.

Server

Swift-on-server

Vapor, Hummingbird, and SwiftNIO in their native environment. First-class Swift 6 concurrency.

The fleet

Latest Apple Silicon. No wait-list.

104-machine fleet as of March 2026: 24× M4 Minis, 48× M4 Pro Minis, 24× M4 Max Minis, 8× M4 Ultra Studios. Most SKUs available within 20 minutes.

Machine CPU GPU RAM SSD Hourly Monthly

M4 Mini 10-core 10-core 24GB 512GB $0.39 $129

M4 Pro Mini 12-core 16-core 48GB 1TB $0.69 $219

M4 Max Mini 16-core 40-core 128GB 4TB $1.19 $379

M4 Ultra Studio 32-core 80-core 256GB 8TB $2.49 $849

All tiers include a 10Gbps port, 5TB of egress per month, and a full-disk reimage between tenants. M4 Ultra Studios have a 2-week lead time when stock is depleted — we'll quote honestly.