$ inference --device=local --cloud=none

Your models.
Your machine.
Your data.
Your software.

Mortise runs open LLMs entirely on your Mac, iPhone, and iPad. You own your data, your software, and your hardware. Private by physics, not by policy.

Coming Soon
Mortise app icon
Mortise chat screen

// 100% on-device

No cloud. No account. No leaks.

Mortise runs curated MLX models directly on Apple Silicon. Your prompts, your documents, and every token generated stay on your device — there is no server to trust because there is no server at all.

Mortise model catalog

// 12 curated models

A model catalog that fits your hardware.

Qwen, Llama, Phi, Gemma, Nemotron, SmolLM3 and more — open-weight models, most licensed Apache 2.0 or MIT. You download the weights onto your own disk and run them free, for as long as you like: nothing metered, nothing rented, no server in the loop. Each entry lists on-device memory estimates so you know what fits your Mac, iPad, or iPhone.

Browse all models →

// local RAG

Chat with your documents.

Attach PDFs and ask questions. Mortise indexes them locally with on-device embeddings and picks the right retrieval strategy — whole document, chunked, or summarized — without your files ever leaving the machine.

Mortise knowledge base

// knowledge bases

Group your PDFs into a searchable library.

Collect related documents into a named knowledge base and search all of them at once. Mortise ranks the matching passages, stitches neighbouring ones back into whole excerpts, and has the model answer with numbered citations pointing at the source file — indexing, ranking, and answering all on your own machine.

Mortise chat with cited sources

// live web search

Grounded answers, cited sources.

When a question needs fresh information, models can call a live web search tool and answer with clickable citations — the search happens on demand, and only the query leaves your device.

How web search works →
Mortise conversation history

// cross-model memory

Conversations that survive model swaps.

Switch from a 1B model to an 8B model mid-conversation and keep the thread. Conversation memory is shared across every model in the catalog, so your context travels with you.

// one app, three platforms

Buy it once. Run it on iPhone, iPad, and Mac.

Mortise is a universal app — one purchase covers all three, with the same catalog and the same features on each. Every install keeps its own encrypted store on the device, so nothing has to round-trip through a cloud to reach your other hardware.

// encrypted at rest

Encrypted at rest, and only your device holds the key.

Conversations, imported PDFs, and everything derived from them are encrypted where they sit — not just tucked behind a login. The key is generated on your device and kept locally in the Apple Keychain, so it never syncs anywhere. On iPhone and iPad, Apple's file protection keeps the files unreadable while the device is locked.

Read the privacy policy →

// one_time_purchase

One purchase. Yours forever.

No subscription, no per-token billing, no account to keep alive. Software development is circular — the app keeps getting better, and updates keep landing on the copy you already own. There is no server of ours in the path, so nothing you paid for can be switched off from our end.

// sovereign_vault_protocol

Enter the Sovereign Vault

Mortise is coming soon. No account required. Nothing leaves your device.

Coming Soon