LocalEngine
Productividad
Gratis · Compras dentro de la app
LocalEngine runs open-source AI language models entirely on your iPhone and iPad. Everything happens on-device: your prompts, conversations, and images never leave your device and are never sent to any server.
Built in pure Swift with SwiftUI, LocalEngine runs models through a Metal-accelerated llama.cpp engine, so you get fast, low-latency AI completely offline — no account, no sign-in, no cloud.
WHAT YOU CAN DO
• Chat — Hold multi-turn conversations with a local model. Replies stream token-by-token in real time. Start a new chat or stop a response at any time.
• See images — Turn on vision in Settings, attach photos to your message, and ask about them. Images are understood entirely on-device by a vision-capable model.
• Download models — Browse a curated catalog of open models and download the one that fits your device. Downloads run as resumable background jobs with live progress, and the screen can stay awake while a large model downloads.
• Tune generation — Adjust the response length (max tokens) and the temperature to shape how the model replies.
• Unlock larger models — LocalEngine is free to use with starter models. A one-time Pro upgrade unlocks larger catalog downloads on iPhone and iPad.
CURATED MODELS
LocalEngine is ready to chat out of the box with a small, fast default model, and its catalog includes compact vision-language models you can grow into:
• Qwen3.5 — 0.8B and 2B free, with 4B available through Pro
• Gemma 4 — E2B free, with E4B available through Pro
Smaller models are quick and light on storage; larger Pro models are more capable on newer devices. Each vision model pairs with a multimodal projector so it can understand images — all on-device.
PRIVATE BY DESIGN
• 100% on-device inference — no cloud, no telemetry, no analytics.
• No account and no sign-in required.
• Runs fully offline once a model is downloaded.
• Built to App Store privacy and security standards.
REQUIREMENTS
• iOS / iPadOS 17.0 or later
• A recent iPhone or iPad is recommended, especially for the larger models
• Free storage space for the models you choose to download
LocalEngine puts the power of modern AI — text and vision — in your pocket, without giving up your privacy.
Más • Updated on-device engine.
• Fixed an issue where LFM and Qwen 3.5 models could carry over text from a previous conversation into a new one.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.7.2 hace 23 h
• Updated on-device engine.
• Fixed an issue where LFM and Qwen 3.5 models could carry over text from a previous conversation into a new one.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.7.1 7 nov
• LFM3 1.7B — Liquid AI's compact multilingual model that thinks before it answers. Free, 1.1 GB download.
• LFM3-VL 2B — ask questions about your photos and images. Free, about 1.3 GB download.
• Updated on-device engine with support for the latest model architectures.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.7.0 29 oct
Version 0.6.8
• Spark X2.6 4B — a compact English and Chinese model for coding, agent and long-context tasks. It thinks before it answers. 3.8 GB download, best on newer iPhone and iPad models.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.6.8 24 oct
• Example prompts — a new chat now offers ready-made prompts for everyday tasks: summarize, translate, rewrite, explain code, draft an email, write meeting minutes. Tap "All prompts" for 20 templates across writing, language, data, coding and work.
• Fill in a short form and start the chat. Long fields can be filled straight from a file, including text, Markdown and PDF.
• Prompts follow the app language, in all 10 supported languages.
• The app warns you before sending when the input is too long for the current model.
• Support iPhone Duo.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.6.7 23 oct
Version 0.6.6
• Example prompts — a new chat now offers ready-made prompts for everyday tasks: summarize, translate, rewrite, explain code, draft an email, write meeting minutes. Tap "All prompts" for 20 templates across writing, language, data, coding and work.
• Fill in a short form and start the chat. Long fields can be filled straight from a file, including text, Markdown and PDF.
• Prompts follow the app language, in all 10 supported languages.
• The app warns you before sending when the input is too long for the current model.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.6.6 21 oct
• Bonsai 2 Small — PrismML's ternary Bonsai 2 models arrive in a new collection in Models: 2B (free), 4B and 9B (Pro), with image understanding and tool calling, at about half the size of regular 4-bit models.
• Updated on-device AI engine for newer model support and fixes.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.6.5 17 oct
• MLX models are here — Qwen3.5, Gemma 4, MiniCPM5 and LFM2.5 now also come in MLX versions, built for Apple silicon. Find them next to the GGUF versions in Models.
• Automatic engine switching — pick any model and LocalEngine uses the right engine for it, no settings to change.
• Updated on-device AI engines for better stability and model support.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.6.3 11 oct
• MLX models are here — Qwen3.5, Gemma 4, MiniCPM5 and LFM2.5 now also come in MLX versions, built for Apple silicon. Find them next to the GGUF versions in Models.
• Automatic engine switching — pick any model and LocalEngine uses the right engine for it, no settings to change.
• Updated on-device AI engines for better stability and model support.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.6.2 8 oct
• Now in 10 languages — the entire interface is available in English, Spanish, French, German, Brazilian Portuguese, Japanese, Korean, Simplified Chinese, Traditional Chinese, and Italian.
• Choose your language in the app — Settings ▸ Language lets you pick a language or follow your device's setting. The change applies the next time you open LocalEngine.
• Fully translated, not just the labels — onboarding, model downloads, chat, projects, and error messages are all localized.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.5.2 12 sep
• Now in 10 languages — the entire interface is available in English, Spanish, French, German, Brazilian Portuguese, Japanese, Korean, Simplified Chinese, Traditional Chinese, and Italian.
• Choose your language in the app — Settings ▸ Language lets you pick a language or follow your device's setting. The change applies the next time you open LocalEngine.
• Fully translated, not just the labels — onboarding, model downloads, chat, projects, and error messages are all localized.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.5.1 5 sep
• Now in 10 languages — the entire interface is available in English, Spanish, French, German, Brazilian Portuguese, Japanese, Korean, Simplified Chinese, Traditional Chinese, and Italian.
• Choose your language in the app — Settings ▸ Language lets you pick a language or follow your device's setting. The change applies the next time you open LocalEngine.
• Fully translated, not just the labels — onboarding, model downloads, chat, projects, and error messages are all localized.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.5.0 26 ago
• New free LFM2.5 2.6B model — download Liquid AI's LFM2.5 2.6B Q4_K_M GGUF model for on-device text generation and tool use.
• Native LFM prompting — when you select an LFM model, LocalEngine now automatically uses the model's built-in chat and tool-call template for more reliable responses.
• Better tool calls — LFM's native tool-call output is recognized and passed into LocalEngine's on-device tool workflow.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.4.20 17 ago
• New free LFM2.5 2.6B model — download Liquid AI's LFM2.5 2.6B Q4_K_M GGUF model for on-device text generation and tool use.
• Native LFM prompting — when you select an LFM model, LocalEngine now automatically uses the model's built-in chat and tool-call template for more reliable responses.
• Better tool calls — LFM's native tool-call output is recognized and passed into LocalEngine's on-device tool workflow.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.4.18 12 ago
Local embeddings — select a dedicated GGUF embedding model independently from your chat model, ready for private semantic search and retrieval.
• New embedding models — download all-MiniLM-L6-v2, BGE Small EN v1.5, or Qwen3 Embedding 0.6B directly on iPhone and iPad.
• Better multilingual retrieval — Qwen3 Embedding 0.6B is recommended for Chinese and other multilingual content.
• Clearer model management — chat and embedding models are now shown separately, so choosing embeddings never interrupts an active chat model.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.4.16 30 jul
• Projects & chat history — a new Projects tab organizes your conversations into projects. Browse, search, rename, and reopen past chats anytime.
• Collapsible projects — expand or collapse each project to keep long histories tidy.
• Quick actions — swipe or long-press a chat to rename, clear, or delete it, and start a new chat or project right from the tab.
• Polish — refreshed chat UI and assorted settings refinements.
0.4.15 23 jul
• Projects & chat history — a new Projects tab organizes your conversations into projects. Browse, search, rename, and reopen past chats anytime.
• Collapsible projects — expand or collapse each project to keep long histories tidy.
• Quick actions — swipe or long-press a chat to rename, clear, or delete it, and start a new chat or project right from the tab.
• Polish — refreshed chat UI and assorted settings refinements.
0.4.14 19 jul
• Projects & chat history — a new Projects tab organizes your conversations into projects. Browse, search, rename, and reopen past chats anytime.
• Collapsible projects — expand or collapse each project to keep long histories tidy.
• Quick actions — swipe or long-press a chat to rename, clear, or delete it, and start a new chat or project right from the tab.
• Polish — refreshed chat UI and assorted settings refinements.
0.4.13 16 jul
• Projects & chat history — a new Projects tab organizes your conversations into projects. Browse, search, rename, and reopen past chats anytime.
• Collapsible projects — expand or collapse each project to keep long histories tidy.
• Quick actions — swipe or long-press a chat to rename, clear, or delete it, and start a new chat or project right from the tab.
• Polish — refreshed chat UI and assorted settings refinements.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.4.12 14 jul
• Projects & chat history — a new Projects tab organizes your conversations into projects. Browse, search, rename, and reopen past chats anytime.
• Collapsible projects — expand or collapse each project to keep long histories tidy.
• Quick actions — swipe or long-press a chat to rename, clear, or delete it, and start a new chat or project right from the tab.
• Polish — refreshed chat UI and assorted settings refinements.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.4.11 9 jul
Version 0.4.10
• Projects & chat history — a new Projects tab organizes your conversations into projects. Browse, search, rename, and reopen past chats anytime.
• Collapsible projects — expand or collapse each project to keep long histories tidy.
• Quick actions — swipe or long-press a chat to rename, clear, or delete it, and start a new chat or project right from the tab.
• Polish — refreshed chat UI and assorted settings refinements.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
0.4.10 8 jul
• Updated on-device engine.
• Fixed an issue where LFM and Qwen 3.5 models could carry over text from a previous conversation into a new one.
Everything still runs 100% on-device — no account, no cloud, no telemetry.
Más Versión 0.7.2 hace 23 h
No se recopilan datos El desarrollador no recopila ningún dato en esta app.