Ensu: The Local LLM That Never Phones Home
While GitHub announces it'll train AI on your private code starting April 24, a small team behind Ente Photos just launched Ensu — a ChatGPT alternative that runs entirely on your device, costs nothing, and never shares your data. In an era where every AI interaction seems to end up in a corporate training dataset, Ensu represents a fundamentally different approach to artificial intelligence.
The Privacy Context: Why Ensu Matters Now
GitHub's recent policy change is just the latest reminder that centralized AI services come with hidden costs. When you use ChatGPT, Claude, or Copilot, your conversations don't stay between you and the model — they fuel future iterations, get reviewed by safety teams, and exist in databases that could be breached or subpoenaed.
Ensu arrives at exactly the right moment. Built by the same team that created Ente Photos — an open-source, end-to-end encrypted alternative to Google Photos — Ensu applies the same privacy-first philosophy to AI. No accounts required. No data collection. No cloud dependency. Just a local LLM that respects your boundaries.
What Ensu Actually Does
At its core, Ensu provides a familiar ChatGPT-style interface for text conversations, but with one crucial difference: everything happens on your device. The model runs locally, processes your inputs locally, and stores your chat history locally. You can use it on flights, in remote areas, or anywhere with zero connectivity.
The app supports image attachments, letting you analyze photos without uploading them to external servers. It comes pre-trained with knowledge of classics like the Gita and Bible, providing thoughtful responses without needing web search capabilities. For developers and privacy-conscious users, this means getting helpful AI assistance without the surveillance.
Technical Architecture
Ensu is built for performance and portability. The core engine is written in Rust, giving it the speed and memory safety needed for efficient on-device inference. Mobile apps are native per platform (iOS and Android), while the desktop versions use Tauri with a shared Rust core, ensuring consistent behavior across operating systems.
The entire project is open source under the MIT license, with full code available on GitHub. This transparency means anyone can audit the code, verify the privacy claims, or fork the project for their own needs. Coming soon is optional end-to-end encrypted sync across devices using an Ente account — meaning you can maintain conversation continuity without compromising privacy.
The Philosophy: AI Autonomy vs. Corporate Control
Ente's manifesto cuts straight to the heart of the matter: "LLMs are too important to be left to big tech." The argument is simple but compelling. Centralized AI creates manipulation risks, exposes users to arbitrary bans, and produces non-portable memory locked inside corporate walled gardens. Your conversations with ChatGPT don't transfer to Claude. Your Copilot history stays with GitHub.
Local AI offers an alternative path: full control, full privacy, and no dependency on companies that might change their policies tomorrow. As the gap between frontier models and local alternatives continues to narrow, the threshold for choosing privacy over convenience gets lower every month.
Comparison to Alternatives
Compared to Ollama — another popular local LLM solution — Ensu focuses on user experience and cross-platform consistency rather than just model management. While Ollama excels at running various models locally, Ensu provides a polished, consumer-ready interface that happens to use local models under the hood.
Against cloud-based alternatives like ChatGPT or Claude, the trade-offs are clear. Ensu won't match the raw capabilities of GPT-4 or Claude 3 Opus — yet. But for many use cases — drafting emails, brainstorming ideas, learning concepts, or coding assistance — local models have reached "good enough" status. The privacy benefits and zero ongoing costs make the capability gap worthwhile for many users.
Current Limitations
Ensu is still early in its journey. The current v1.0 release focuses on text conversations with image attachments, but lacks some features power users might expect. There's no web browsing capability, no plugin ecosystem, and the available models won't match the latest frontier models from OpenAI or Anthropic.
The cross-device sync feature is still in development, so for now your conversations stay on the device where they started. And while the app supports iOS, Android, macOS, Linux, and Windows plus an experimental web version, performance will vary significantly based on your device's hardware capabilities.
FAQ
Is Ensu completely free?
Yes. No subscriptions, no API keys, no usage limits. Because the AI runs entirely on your device, there are no server costs for Ente to pass on to users. The project is funded by Ente's other products and community support.
What models does Ensu use?
Ensu uses locally-run open-source language models. The specific models bundled may vary by platform and version, but they're chosen to balance capability with resource usage on consumer hardware.
Can I use my own models?
The current v1.0 release comes with bundled models optimized for the Ensu experience. Future versions may expand model flexibility, but the focus is on providing a polished out-of-the-box experience rather than model tinkering.
How does this compare to running Ollama locally?
Ollama is designed for developers who want to run and switch between various models. Ensu is designed for end users who want a ChatGPT-like experience that happens to be private. Think of Ollama as the engine and Ensu as the consumer vehicle.
Is the end-to-end encrypted sync available now?
Not yet — it's listed as coming soon. For now, conversations are stored locally on each device. When sync launches, it will use Ente's end-to-end encryption infrastructure.
What about enterprise or team use?
Ensu is currently positioned as a personal productivity tool. The self-hostable sync infrastructure may appeal to privacy-conscious organizations, but there's no dedicated enterprise tier or admin controls at this stage.
