CuriousLM

About CuriousLM

CuriousLM is a free AI app that runs the model on your phone or computer, so chats are written and stored on the device, encrypted. There's no account and no chat server, and it keeps working in airplane mode. Every model it runs comes straight from Hugging Face, and the default is Curious 2B, a roughly 1.2 GB download that happens once inside the app.

Review

CuriousLM launched this week and takes a local-first approach to AI chat. The app ships with a journal, on-device voice, and Pocket Wikipedia, a 60 MB copy of Wikipedia's 121,797 most read articles that answers cite. The maker lists honest limits: small models are slower and less clever than the giant cloud ones, and the app is best on devices with 8 GB of RAM or more.

Key Features

• Heat. Words arrive hot and cool to ink, and the ones the model was unsure of stay underlined. Tap an underlined word to see what else it nearly said and take the answer that way instead.
• On-device chat. The model runs on your phone or laptop and chats stay there, encrypted. There's no account and nothing to sign up for.
• It speaks first (Android). With your permission, source by source, the app reads your notifications and calendar, then tells you who is waiting on a reply and what's due. Processing happens on the device.
• Voice, Pocket Wikipedia and a journal. All three run on the device, and Wikipedia answers cite the stored articles.

Pricing and Value

CuriousLM is listed as free. The reference content gives no paid tiers, subscriptions, or usage limits, and pricing beyond the free app is not yet defined. The cost you carry is storage and hardware: the default Curious 2B model takes about 1.2 GB on disk, and the maker says the app works best on devices with 8 GB of RAM or more.

Pros

• No account and no chat server. Chats stay on the device, encrypted, and the app keeps working in airplane mode.
• Heat shows where the model was choosing between words, so you can see the other paths it nearly took. The maker is clear that this isn't fact checking.
• Pocket Wikipedia fits 121,797 articles in 60 MB, and answers cite them without a round trip.
• The Android "speaks first" mode reads notifications and calendar with your permission, source by source, and tells you who is waiting on a reply.

Cons

• Small models are slower and less clever than the giant cloud ones, per the maker's own limits. A 2B model won't match what people get from cloud chatbots.
• It's best on devices with 8 GB of RAM or more. A user on this page asked whether the 2B model holds up below that, and no answer was posted.
• Not well suited to people who need top-end answer quality or heavy multitasking across long documents. The device-first design trades capability for privacy.

CuriousLM fits people who want to ask an AI personal questions, paste a message from a family member, or let it read a calendar without that content reaching someone else's server. The 8 GB RAM requirement and the 2B model's limits mean it suits lighter, privacy-sensitive use rather than work that depends on the strongest available answers.