QuantLM - Offline AI Chat

Mohammad Yaser Hussain

About this app

Most AI apps are really just a chat window bolted onto someone else's server. Type a question, it travels across the internet, gets answered by a model you'll never see, and who knows what happens to it after that. QuantLM works differently. The model lives on your phone, the thinking happens on your phone, and once you've downloaded it, none of that needs an internet connection at all.

𝘘𝘶𝘢𝘯𝘵𝘓𝘔 𝘪𝘴 𝘴𝘵𝘪𝘭𝘭 𝘢𝘤𝘵𝘪𝘷𝘦𝘭𝘺 𝘣𝘦𝘪𝘯𝘨 𝘣𝘶𝘪𝘭𝘵. 𝘕𝘦𝘸 𝘮𝘰𝘥𝘦𝘭𝘴 𝘢𝘯𝘥 𝘧𝘦𝘢𝘵𝘶𝘳𝘦𝘴 𝘢𝘳𝘦 𝘢𝘥𝘥𝘦𝘥 𝘳𝘦𝘨𝘶𝘭𝘢𝘳𝘭𝘺, 𝘢𝘯𝘥 𝘸𝘩𝘪𝘭𝘦 𝘵𝘩𝘦 𝘤𝘰𝘳𝘦 𝘢𝘱𝘱 𝘪𝘴 𝘴𝘵𝘢𝘣𝘭𝘦, 𝘺𝘰𝘶 𝘮𝘢𝘺 𝘳𝘶𝘯 𝘪𝘯𝘵𝘰 𝘵𝘩𝘦 𝘰𝘤𝘤𝘢𝘴𝘪𝘰𝘯𝘢𝘭 𝘳𝘰𝘶𝘨𝘩 𝘦𝘥𝘨𝘦, 𝘦𝘴𝘱𝘦𝘤𝘪𝘢𝘭𝘭𝘺 𝘰𝘯 𝘭𝘦𝘴𝘴 𝘤𝘰𝘮𝘮𝘰𝘯 𝘥𝘦𝘷𝘪𝘤𝘦𝘴. 𝘍𝘰𝘶𝘯𝘥 𝘴𝘰𝘮𝘦𝘵𝘩𝘪𝘯𝘨 𝘣𝘳𝘰𝘬𝘦𝘯? 𝘛𝘩𝘢𝘵 𝘧𝘦𝘦𝘥𝘣𝘢𝘤𝘬 𝘨𝘰𝘦𝘴 𝘴𝘵𝘳𝘢𝘪𝘨𝘩𝘵 𝘪𝘯𝘵𝘰 𝘵𝘩𝘦 𝘯𝘦𝘹𝘵 𝘶𝘱𝘥𝘢𝘵𝘦..

𝗖𝗔𝗡 𝗬𝗢𝗨𝗥 𝗣𝗛𝗢𝗡𝗘 𝗛𝗔𝗡𝗗𝗟𝗘 𝗜𝗧

Local AI is a memory game more than anything else. Smaller models get by comfortably on around 4GB of RAM. The stronger ones start wanting 8GB, and the biggest models in the library ask for as much as 12GB. Rather than make you guess, QuantLM lists the exact requirement on every model before you tap download, so you only pull down what your phone can actually run.

𝗘𝗩𝗘𝗥𝗬𝗧𝗛𝗜𝗡𝗚 𝗜𝗡𝗦𝗜𝗗𝗘

- 𝗡𝗮𝘁𝘂𝗿𝗮𝗹 𝗖𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻𝘀: answers stream in live, and your history is saved so you can pick a thread back up days later.
- 𝗖𝘂𝗿𝗮𝘁𝗲𝗱 𝗠𝗼𝗱𝗲𝗹 𝗟𝗶𝗯𝗿𝗮𝗿𝘆: a handpicked set of models from Google, Microsoft, Alibaba, and Hugging Face, so there's a real choice between fast and lightweight or slower and sharper.
- 𝗕𝗿𝗶𝗻𝗴 𝗬𝗼𝘂𝗿 𝗢𝘄𝗻 𝗠𝗼𝗱𝗲𝗹: already have a model file you'd rather use? Load it straight in.
- 𝗩𝗶𝘀𝗶𝗼𝗻: hand it a photo and have it describe, read aloud, break down, identify, or translate whatever's in frame.
- 𝗩𝗼𝗶𝗰𝗲 𝗜𝗻, 𝗩𝗼𝗶𝗰𝗲 𝗢𝘂𝘁: ask out loud and get the answer read back to you, and a few models can even listen to short audio clips on their own.
- 𝗦𝗸𝗶𝗹𝗹𝘀: teach the assistant new tricks without the app ever reaching outside the model running on your phone.
- 𝗧𝘂𝗻𝗮𝗯𝗹𝗲 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻 𝗦𝗲𝘁𝘁𝗶𝗻𝗴𝘀: adjust temperature, system prompt, and other parameters yourself, or just leave the defaults alone.
- 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗥𝗲𝗮𝗱𝗼𝘂𝘁: a running tokens per second count so you can see exactly how a model performs on your phone.
- 𝗔𝗽𝗽 𝗟𝗼𝗰𝗸: PIN, password, pattern, or fingerprint.
- 𝗡𝗼 𝗔𝗰𝗰𝗼𝘂𝗻𝘁 𝗥𝗲𝗾𝘂𝗶𝗿𝗲𝗱: nothing to sign up for, no account screen between you and the first message.
- 𝗢𝗽𝘁𝗶𝗼𝗻𝗮𝗹 𝗪𝗲𝗯 𝗟𝗼𝗼𝗸𝘂𝗽: the one part of QuantLM that can reach the internet. Switch it on and the assistant pulls in fresh results when a question genuinely needs something current. Leave it off and the app never touches the internet at all.

𝗧𝗛𝗘 𝗛𝗢𝗡𝗘𝗦𝗧 𝗧𝗥𝗔𝗗𝗘 𝗢𝗙𝗙

QuantLM won't out argue a massive model running across a data center somewhere. It can't. What fits on a phone is smaller than what fits in a server rack, and that's just physics. What you get instead is a model that's entirely yours while you're using it. Nothing typed into this app sits on a company's servers, nothing needs a strong signal to keep working, and nothing about how it behaves was decided by anyone but you.

𝗜𝗦 𝗧𝗛𝗜𝗦 𝗧𝗛𝗘 𝗥𝗜𝗚𝗛𝗧 𝗔𝗣𝗣 𝗙𝗢𝗥 𝗬𝗢𝗨

If your priority is squeezing out the smartest possible answer no matter where your conversation ends up, there are excellent cloud assistants built exactly for that.

If you'd rather keep your questions between you and your phone, and you don't mind trading a bit of raw power for that, QuantLM was built for exactly this.

What’s new

• Improved photo/video picker for better privacy. Only access the media you select
• More reliable model loading and download handling
• Fixed edge-case crashes and improved app stability
• Performance and storage optimizations