19 families · one private engine
Find the local model with the right instinct.
Every download runs on your Apple hardware. Start with a quick conversational model, bring an image into the loop, give an agent a tool, or build a private semantic index—without turning your work into someone else’s data.
Small enough to travel.
Model families with downloads configured for the LocalEngine iOS catalog. Some are shared with Mac; their detail page compares both.
Qwen 3.5
Qwen
0.8B · 2B · 4B
The pocket-to-workstation generalist: a single family that can read a screenshot, hold a conversation, and hand structured work to a tool without sending the moment away from your device.
Explore the familyGemma 4
E2B · E4B
A calm, image-literate lineup for turning what is on the screen into useful language—captioning, explaining, comparing, and drafting without an upload step.
Explore the familyQwen 3.8
Qwen
2B · 4B
A reasoning-first branch that feels at home in a plan: it can turn an instruction into a sequence, a function call, or—in its larger form—a vision-aware answer.
Explore the familyBonsai 2 Small
PrismML
2B · 4B · 9B
A pocket garden of ternary vision models: the small Bonsai line pares down the weight format so an iPhone or iPad can keep image-aware chat close at hand.
Explore the familyLFM 2.5
Liquid AI
LFM2.5 2.6B · LFM2.5-VL 1.6B
A light-footed on-device pair: one for text agents that think and call tools, another for turning a camera roll, a label, or a sign into an offline conversation.
Explore the familyMiniCPM5
OpenBMB
1B
A tiny practicalist: the model to keep ready when the useful answer is a short, private text exchange or a tool action—not a heavyweight production.
Explore the familyEmbeddings
Multiple authors
Qwen3 Embedding 0.6B · Nomic Embed Text v1.5 · BGE Small EN v1.5 · all-MiniLM-L6-v2
The quiet index behind an intelligent library: these models turn notes, documents, and snippets into vectors so a local app can find meaning rather than merely match words.
Explore the familyA deeper local bench.
From compact generalists to high-memory workstation models, all backed by LocalEngine’s macOS catalog.
Qwen 3.5
Qwen
0.8B · 4B · 9B · 27B · 35B-A3B
The pocket-to-workstation generalist: a single family that can read a screenshot, hold a conversation, and hand structured work to a tool without sending the moment away from your device.
Explore the familyGemma 4
E2B · E4B · 12B · 26B-A4B · 31B
A calm, image-literate lineup for turning what is on the screen into useful language—captioning, explaining, comparing, and drafting without an upload step.
Explore the familyQwen 3.8
Qwen
2B · 4B · 27B
A reasoning-first branch that feels at home in a plan: it can turn an instruction into a sequence, a function call, or—in its larger form—a vision-aware answer.
Explore the familyQwen 3.6
Qwen
27B · 35B-A3B
A Mac-scale visual reasoning line that trades the smallest sizes for more room to read an image, work through a task, and pass the result into a tool-shaped workflow.
Explore the familyBonsai 2
PrismML
27B
The small-tree idea made literal: ternary weights squeeze a vision-and-reasoning family into downloads that leave more of your device free for the rest of your day.
Explore the familyOrnith 1.0
DeepReinforce AI
9B · 35B
A codebench companion with the habits of an agent: it reads the assignment, uses a tool, and keeps moving through structured work rather than only narrating it.
Explore the familyOrnith 1.5
Ornith AI
9B · 35B-A3B · 397B
The taller Ornith: a reasoning-and-coding flock that adds vision, native tools, and a far-reaching flagship for Macs built to carry it.
Explore the familyLaguna XS 2.1
Poolside
33B total · 3B active
A compact-looking deep-water coder: 33B total parameters, only 3B active per token, and enough patience for the work that unfolds over many steps.
Explore the familyLaguna S 2.1
Poolside
118B total · 8B active
The long-horizon heavyweight: an 118B mixture-of-experts coding model that turns a well-provisioned Mac into a private workshop for stubborn technical problems.
Explore the familyDeepSeek V4 Flash
DeepSeek
284B
A 284B agentic engine for the rare Mac that can host a private, large-scale thinking partner—five shards of weight for a workspace that refuses to leave the workstation.
Explore the familyQwen AgentWorld
Qwen
35B-A3B
A language world model for the rehearsal before the action: it can simulate an agentic environment so plans can be tested, revisited, and improved locally.
Explore the familyQwythos
Empero AI
9B v2
A vision-language model with a multi-token rhythm: it is built to keep an image-and-text conversation moving with an eye toward responsive local interaction.
Explore the familyMuse Glimmer
Meta Models
30B
A 30B multimodal agent that arrives with its perception encoder—the visual sensor and the reasoning engine travelling together in one local kit.
Explore the familyClef
Cloudflare
Clef Flash 9B · Clef 27B
Not a conversationalist but a decisive specialist: ask for a choice, score, classification, or extraction and receive a typed answer meant for a system to act on.
Explore the familyEmbeddings
Multiple authors
Qwen3 Embedding 0.6B · Nomic Embed Text v1.5 · BGE Small EN v1.5 · all-MiniLM-L6-v2
The quiet index behind an intelligent library: these models turn notes, documents, and snippets into vectors so a local app can find meaning rather than merely match words.
Explore the familyQwen Image
Qwen
Qwen-Image 20B
A private image studio inside the engine: describe a scene in Chinese or English and let the Mac compose it locally, including text that often survives the trip into pixels.
Explore the familyPick the job before the parameter count.
The best local model is the smallest one that can do your work comfortably. Keep a compact model for everyday questions and add a specialist only when its capability earns the storage.
Need a first model? Begin with Qwen 3.5 0.8B on phone or Mac.
Need to see? Choose a family with the Vision badge—Qwen, Gemma, Bonsai, LFM, and others.
Need to find? Add Embeddings for private semantic search instead of using a chat model as an index.