What Happened
Ente, the company known for its end-to-end encrypted photo storage and authenticator apps, released Ensu on March 2, 2026. It is a ChatGPT-style chat app that runs large language models entirely on your device with no cloud processing.
Ensu ships with two quantized models: LFM 2.5 VL 1.6B (about 664 MB, the default) and Gemma 3 4B IT for higher-RAM macOS machines. Both run on CPU only. The app supports text and image inputs, so you can describe photos or ask questions about screenshots without anything leaving your phone or laptop.
Platform support is broad: iOS, Android, macOS (both M-series and Intel), Windows, Linux, and even web browsers. The core logic is written in Rust, with native mobile wrappers and Tauri-based desktop apps sharing the same backend. The whole project is open source under Ente's GitHub repository.
There are real constraints. Context length tops out at 4,096 tokens with a 512 token output cap. There is no web search, no OCR, and no audio transcription. Ente plans to add encrypted cross-device chat sync using your Ente account (or self-hosted instance), but that is not available yet.
The app is free. Ente calls it a "Labs project," meaning they are iterating on direction before committing to pricing or stability guarantees.
Why It Matters
The market for local LLM apps is getting crowded, but most options still require some technical setup. Ollama needs command-line comfort. LM Studio targets power users. Ensu takes a different approach: download, pick a model, start chatting. No Docker, no API keys, no config files.
For anyone handling sensitive data - legal documents, medical notes, financial records - local-only processing removes the trust question entirely. Nothing hits a server. Period. That is a meaningful guarantee that cloud-based alternatives simply cannot match, regardless of their privacy policies.
The image input support on a sub-700 MB model is worth noting. Running multimodal inference on a phone with no internet connection was science fiction two years ago. The quality will not rival GPT-4o or Claude, but for quick tasks like reading a receipt or summarizing a whiteboard photo, it fills a gap.
Our Take
Ensu is not competing with ChatGPT or Claude on capability. A 1.6B parameter model with 512 output tokens is not going to write your quarterly report. But that is not the point.
What Ente is building is a privacy-first utility layer. The same way their Photos app offers a Google Photos alternative for people who care about encryption, Ensu offers a ChatGPT alternative for people who refuse to send their data to someone else's server. If you already use Ente's ecosystem, the upcoming encrypted sync will make this a natural addition.
The Rust core shared across all platforms is a smart architectural choice. It means updates ship everywhere at once, and the performance overhead stays low on mobile. This is the kind of engineering decision that matters for long-term maintenance.
The 4K context limit is the real bottleneck. You cannot feed it a long document or have extended conversations without hitting the wall. If Ente can raise that ceiling and add more capable models, Ensu could become a default recommendation for privacy-conscious users. Right now, it is a solid proof of concept with a clear trajectory.