Typical cloud AI
Rented intelligence.
- Your prompts are processed on someone else’s servers
- No signal, no answers
- Account, sign-in and a monthly subscription
- One vendor-controlled model
- Trust is “read our privacy policy”
Open source Offline AI for iPhone · MIT licensed
Your mind.Your device.No cloud.
LokalMind runs open-weight AI models directly on your iPhone. Chat on a plane, in a basement, anywhere — your conversations never leave your device.
- No account
- No telemetry
- Works in airplane mode
Why LokalMind
Most AI apps send everything you type to a server.
LokalMind flips the model. The AI lives on your phone, answers are generated on your phone’s chip, and nothing is transmitted anywhere. Privacy isn’t a policy page — it’s the architecture.
LokalMind
Intelligence you own.
- Inference runs entirely on your iPhone
- Works fully offline once a model is downloaded
- No account, no API key, no per-token meter
- Pick from a catalog of open-weight models
- Trust is architectural — and the code is open
FeaturesA real AI assistant.
A real AI assistant.
Just not in the cloud.
Everything you expect from a modern chat app — streaming answers, memory, modes, search — rebuilt to run locally with a premium, native iOS feel.
Streaming replies, generated on your chip
Tokens stream in live with haptic feedback. Stop mid-answer, switch models from the chat header, and keep going.
Memory that stays local
Pinned facts, a learned profile and cross-session memories — retrieved with on-device semantic search.
Modes & your own prompts
Switch between General, Coding, Creative and Marketing — and rewrite the system prompt for each one.
Message queue
Keep typing while the model answers. Queued messages sit in a strip — edit or remove them before they send.
Resumable downloads
Large models download in the background — pause, resume and cancel without starting over.
Edit & rewind
Long-press any message you sent to fix it. The thread rewinds from that point and the assistant regenerates from your corrected text.
Code blocks & reasoning
Markdown rendering with language-labelled code blocks and one-tap copy. Chain-of-thought is tucked away so you read the answer first.
Search everything
Find old threads across chat titles and message text. Auto-titles keep your recents tidy.
Sized for your phone
Onboarding matches models to your device’s RAM, so you never download something that can’t run.
Erase in one tap
Wipe every chat, memory, setting and model from the device. A true clean slate.
Take the tour
Designed to feel at home on iOS.
Dark, glass-style surfaces, native tab bar, haptics on the moments that matter. Not a web wrapper — a real iPhone app.
How it works
Private AI in three steps.
1.
Get the app
Download LokalMind from the App Store. No sign-up, no email, no account — just your name, if you like.
2.
Pick a model
Choose an open-weight model sized for your iPhone’s memory. It downloads once and lives on your device.
3.
Chat anywhere
Flip on airplane mode if you want. Every answer is generated locally, and every chat stays on your phone.
Models
Choose your brain.
A curated catalog of open-weight GGUF models from Qwen, DeepSeek and Google. Small ones for speed, bigger ones for depth — switch any time.
Qwen 3.5 · 0.8B
Alibaba · QwenBlazing-fast everyday chat with thinking mode built in. The perfect first model.
550 MB2 GB+ RAMFastest
Qwen 3.5 · 2B
Alibaba · QwenThe balance of speed and intelligence. Strong at summaries and multilingual chat.
1.3 GB3 GB+ RAM
DeepSeek R1 · 1.5B
DeepSeekThinks before it answers. Built for coding, math and step-by-step logic.
1.1 GB3 GB+ RAM
Qwen 3.5 · 4B
Alibaba · QwenMore headroom for complex tasks, long-form writing and creative work.
2.9 GB4 GB+ RAM
Qwen 3 · 4B
Alibaba · QwenThinking mode for harder problems — a strong pick for coding sessions.
2.5 GB4 GB+ RAM
Gemma 3 · 4B
Google · GemmaLong context and careful analysis. Great for working through documents and ideas.
2.5 GB4 GB+ RAM
Tip: start with Qwen 3.5 0.8B for speed, then step up to a 4B model when you need deeper reasoning.
Privacy by architecture
of your chats ever leave your phone.
No accounts. No tracking. No ads. No cloud inference. Your conversations are stored locally, processed locally, and deleted locally.
Read the full privacy policy — it’s short, because there isn’t much to say.
- Your chats & history Stays on device
- Memories & learned profile Stays on device
- System prompts & settings Stays on device
- Analytics & telemetry None
- Account required Never
- Network used for Model downloads & updates only
Open source
Don’t trust us. Read the code.
LokalMind is MIT-licensed and built in public. Every line that touches your data is on GitHub — audit it, fork it, or send a pull request.
Built on a clean, native stack
Expo and React Native with llama.cpp inference via llama.rn, SQLite storage and a Clean Architecture core — domain logic with zero third-party imports.
# run it yourself
$ git clone https://github.com/Sandipan006/lokalmind-app
$ cd lokalmind-app && npm install
$ npx expo run:ios
Expo 54React Native 0.81llama.rnSQLiteMobXMIT
Made by one indie developer
LokalMind is an independent project — no corporation, no investors, no data business. If it’s useful to you, a star or a coffee genuinely keeps it going.
Coming soon
Voice input
Hold-to-talk with on-device Whisper speech recognition.
Planned
Export chats
Share any conversation as PDF or Markdown.
Planned
Tool use
Reminders and Calendar, handled locally via EventKit.
Later
Android
The same private AI, on the other half of the world’s phones.
Do I need internet to chat?
No. Once a model is downloaded, LokalMind works completely offline. The internet is only used to download models and receive app updates.
Where is my data stored?
Everything — chats, memories, settings and models — lives in local storage on your iPhone. Nothing is sent to any server, and there’s no account to sync it to.
Which iPhone do I need?
LokalMind checks your device’s memory during setup and recommends models that fit. Small models like Qwen 3.5 0.8B run on phones with 2 GB+ of RAM; 4B models want 4 GB or more.
How good are on-device models?
Surprisingly capable for everyday writing, summaries, brainstorming, coding help and reasoning. They won’t match the largest frontier models in the cloud — LokalMind trades a bit of raw power for complete privacy and offline access.
Why is the app slow on my device?
Larger 4B models need more RAM and compute. Try a smaller model like Qwen 3.5 0.8B for faster responses on older devices.
How do I reset everything?
Profile → Erase All Data. This wipes all chats, settings, memories and downloaded models from your device and returns you to onboarding.
Take your AI off the grid.
Free on the App Store. No account. No cloud. Just you and a model that lives on your phone.