Current platform boundary
Studio is built for Apple Silicon and a minimum deployment target of macOS 14. The complete native workflow has so far been measured on an Apple M4 Max with 128 GB unified memory. Representative lower-memory Macs remain in the clean-device matrix, so an exact public minimum is not yet approved.
| Area | First-beta boundary | Evidence |
|---|---|---|
| Processor | Apple Silicon, arm64 only | Qualified on Apple M4 Max |
| Operating system | macOS 14 or newer | Deployment target implemented; full-workflow evidence currently on macOS 26.x |
| Unified memory | Public minimum pending | Do not infer a minimum from one high-memory qualification host |
| Local storage | About 755 MB app, 2.3 GB model, at least 5 GB free before download | Measured app/model size and tested disk preflight |
| Internet | First model download and optional update checks only | Anonymous model request and updater endpoint captured |
| Developer tools | Not required by the distributed app | Relocatable Python/MLX engine passed; notarized artifact pending |
The one first-beta workflow candidate
- Model:
mlx-community/Qwen3-4B-Instruct-2507-4bitat pinned revision50d427756c6b1b2fe0c0a10f67fbda1fc8e82c1b; model-card license metadata is Apache-2.0. - Data: UTF-8
train.jsonl,valid.jsonl, andtest.jsonl, with one object per line and stringpromptandanswerfields. - Method: supervised LoRA/QLoRA through the bundled MLX runtime, with bounded Guided or Expert parameters.
- Evidence: dataset health, frozen baseline, candidate evaluation against the same test split, run state, source revisions, and artifact hashes.
- Export: a content-addressed LoRA adapter package with experiment, evaluation, and provenance evidence—not a fused model.
A project may remain open beside other projects, but only one operation runs inside each project. Concurrent training across projects has no public support guarantee and depends on available unified memory.
Explicitly not supported in the first beta
- Intel Macs, Windows, Linux, iPhone, iPad, or virtualized macOS.
- Generic document extraction, OCR, image/audio/video training, general chat, or agent workflows.
- Fused full-model, GGUF, Ollama, Core ML, or Hugging Face publishing export.
- Accounts, remote project storage, managed cloud compute, Multi-Mac execution, BYOC, billing, team roles, or sharing.
- Product analytics, automatic crash reports, background diagnostics, or a diagnostic upload service.
- A guarantee that fine-tuning improves quality or that an adapter works outside its recorded base-model/runtime boundary.
Start with the first incomplete stage
| Problem | Check | Safe action |
|---|---|---|
| Model will not download | Internet access and at least 5 GB free | Retry from Baseline; verified existing caches are preserved |
| Baseline is unavailable | Data and model readiness | Return to the first incomplete stage shown in the journey |
| Training was interrupted | Restored run state | Use Resume only when Studio offers it; otherwise start a new run |
| Mac is under memory pressure | Other ML work and concurrent runs | Stop other workloads and run projects sequentially |
| Export is disabled | Completed run, candidate evaluation, and dataset revision | Complete the prerequisite named by Studio |
| Update check fails | Network and updater setting | Keep the installed version; never bypass signature checks |
A support request should not become a data export
Include the Studio version/build, macOS version, chip and coarse memory tier, rounded free-disk range, workflow stage, exact visible error title/message and any displayed code, and concise reproduction steps. Crop or redact screenshots. Review every attachment.
Studio has no automatic report or upload path. The local engine's developer-facing doctor command can create a redacted bundle from explicitly selected JSON/YAML, but it is not a public in-app send feature. The first beta sends no product telemetry or automatic crash report.
What remains before this matrix is operative
- Developer ID sign and notarize two application versions.
- Complete install, first launch, workflow, recovery, export, update, and uninstall on clean supported accounts.
- Test representative lower-memory Macs and publish the smallest passing unified-memory tier.
- Repeat the complete anonymous model download and network capture from the release account.
- Approve the release legal text and third-party notices.
Need help with a development build?
Send a minimal reproduction without private workload data. Security vulnerabilities use the dedicated security address.