Scan this Mac
We read only the Apple chip, unified memory, storage, macOS version, and Metal support needed for honest compatibility.
ModelSentry reads your Mac, recommends models that truly fit, and handles download, verification, setup, and first run—privately on your computer.
ModelSentry turns the fragile parts of local AI into a guided, trustworthy flow.
We read only the Apple chip, unified memory, storage, macOS version, and Metal support needed for honest compatibility.
Compare lightweight, everyday, coding, reasoning, and quality options with clear fit and speed guidance.
Download, resume, verify, install, warm up, and health-check—handled end to end with automatic retry.
ModelSentry reserves memory for macOS, checks real storage headroom, and tells you plainly when a model is a great fit, likely to run slowly, or should not be installed.
Local inference keeps supported prompts and text documents on the computer. History is optional, local-only, and off by default.
ModelSentry refuses to run files it cannot verify, then keeps checking after installation.
Interrupted transfers pick up where they left off, with bounded retries for temporary failures.
Every downloaded model must match its publisher-controlled catalog digest before installation.
Missing or changed files become Repair Needed instead of creating a mysterious failure later.
The selected model loads and generates locally before ModelSentry calls it Ready.
One Mac app. One-time purchase. Bring the models that fit your machine—ModelSentry handles the hard parts.
Requires Apple silicon and macOS 14 or later. Model downloads use additional storage and internet bandwidth.
Models are optional downloads you choose after Smart Match. ModelSentry 1.0 offers eight publisher-controlled, checksum-pinned GGUF options under Apache-2.0 or MIT terms.
No. ModelSentry 1.0 requires Apple silicon, macOS 14 or later, and Metal. Unsupported Macs stop before model setup.
Model catalog checks and model downloads use the internet. Supported prompts and text documents are processed locally and are not sent to Trieflow.
Yes. Choose up to two during setup, then manage installed models in the Library. Smart Match prevents choices that do not safely fit.