Stackintosh
Your Mac has connections.
Your next idea shouldn’t wait for a test suite. Send reviewed checks to spare Macs. Keep your editing Mac for what’s next.
In development · Apple silicon · macOS 26 or later
Think outside the box.
Choose a reviewed check. Save your files. Let another Mac run it. Every result records exactly which saved files it checked.
One check. A clear path.
You choose the check. Stackintosh moves the saved files. The result keeps the record.
Workflow illustration · Each check runs on one worker.
01 / Choose
Choose a named check from a configuration you reviewed. The executable, arguments and environment are explicit. Stackintosh runs that profile; the source project never chooses the command.
Type checks, linters, tests. The tools you already use.
02 / Send
Save your files. Stackintosh captures the saved work—including uncommitted changes—and sends the snapshot objects a configured worker needs over SSH. The check gets its own workspace. You keep yours.
Your worktree stays yours. Unsaved editor buffers aren’t included.
03 / Follow
Follow the output in the app or terminal. Each result identifies the saved files it checked. You and your coding agents can follow the same run, and closing a client doesn’t stop accepted work.
Later edits don’t rewrite the result. Save and run again to check them.
Spare capacity. On your terms.
Workers check their available slots, memory and current conditions before admitting work. When a required reading is unknown, the check waits.
Heavy checks stay off the editing Mac by default. You can explicitly approve one queued check to run there, one at a time.
Fresh capacity readings and your configured power, thermal and memory rules help decide when work can start.
A worker going quiet doesn’t prove its check has stopped. Stackintosh marks the attempt unknown until it can reconcile what happened.
One check, one worker. Stackintosh does not pool memory, move running processes or distribute an AI model.
AI workflows. A hardware-aware stack.
The right work deserves the right hardware. See the NVIDIA, AMD, Apple and Intel GPUs your workers report, the drivers behind them, and the capacity available for your reviewed checks.
Conceptual silicon illustrations. Hardware and driver reporting depend on the host.
Illustration · Matching a GPU to one reviewed check.
See the GPU kind and the drivers the worker reports. Installed and serving are different states.
See the GPU, its name, and every driver the worker detects—including which one is serving it. Metal appears when detected. Tiny CPU readiness depends on a supported architecture and an available clang or LLVM toolchain. Driver installation stays with you.
Your coding agents can run the same reviewed checks you do. A profile can request a GPU kind and, optionally, a serving driver. Stackintosh reserves a matching device. If none is available, that check waits while other eligible work can continue.
Illustration · Independent checks, each running on one worker.
Start with a spare Mac. Configure up to 255 remote workers. Stackintosh tracks capacity and routes independent checks to available workers. Each check runs on one worker, using that worker’s CPU and memory.
Workers must match the coordinator’s operating system and architecture. GPU detection alone doesn’t establish workload or compute-backend compatibility.
At home on your Mac.
A native Mac app for projects, reviewed checks, workers and results. A command line for your terminal and coding agents. One service keeps the work together.
The app, command line and Swift clients talk to the same service. Your checks can keep running when a client closes.
On supported Macs, request an Apple Intelligence summary of finished output. Processing stays on that Mac; the original log and recorded result remain authoritative.
Create custom Mac shortcuts for Run Reviewed Check, Check Status and Show Workers, then ask Siri to run them by name.
A little setup. More breathing room.
Stackintosh is in development. Start with the source and setup guide; a public app download is not available here yet.
Use an editing Mac and one or more compatible workers. Install the matching service and the tools your checks need on each worker.
Verify SSH host identity and name your workers in macbridge.toml. Bonjour can find nearby candidates; pairing is configured manually.
Register a Git project, review its allowed check profiles, save your files and submit a check. Follow the output while you keep working.
Each check runs on one worker with that worker’s CPU and memory. Stackintosh does not combine memory, migrate a running process or move your IDE’s language server. Distributed model inference is a separate future plan, not a current feature.
No. Checks use saved files on disk, including staged and unstaged changes and untracked files Git does not ignore. Unsaved buffers aren’t included. Ignored inputs are included only when the reviewed task profile explicitly lists them. Unsupported repository layouts are refused.
Yes, depending on the reviewed command and its environment. A command, dependency or tool can make its own network requests. Use trusted workers and profiles, and review what your tools can access. Stackintosh does not make an arbitrary command a network-isolated sandbox.
Connected describes the app’s connection to the service on the same Mac. The separate Ready status in Workers describes a worker’s availability. Confirm the configured host, SSH trust, matching service, tools and capacity readings. A nearby Mac is only a candidate until configured.
Open Settings and check Background operation. Enable it if needed, approve any macOS request in System Settings, then choose Check again. Folder allowances for projects, tools and SSH key files apply when the service next starts. Let accepted work finish before restarting.
Snapshots stay in the service’s local state and are transferred over SSH to workers you configure. Check results and output are retained there according to configuration. The current app has no developer-operated source upload or analytics service. Review the privacy draft for data flows, retention and your choices.
A human on the other end.
Need a hand with setup or a check that won’t start? Send your app and macOS versions, the status you see and the relevant error.
Email BrandonPlease remove private source code, passwords, SSH keys and tokens before sending.