Top AI Product

Every major AI assistant sends your conversations to a server. ChatGPT, Gemini, Claude, Copilot — they all require an internet connection and a user account, and your prompts travel through infrastructure you don’t control. For most people, that tradeoff is fine. For a growing subset of users, it’s a dealbreaker.

Ensu is a new local LLM app from Ente, the company behind the end-to-end encrypted photo storage service that passed a CERN-sponsored security audit. It runs entirely offline on your device. No account required. No data leaves your phone or laptop. No usage limits. No cost.

The app hit the Hacker News front page on March 25 with 328 points and 147 comments, triggering a debate that says as much about the state of local AI as it does about Ensu itself.

Who Is Ente and Why Are They Building an AI App?

Ente was founded by Vishnu Mohandas, who left his job and moved back to India during the pandemic to build what he wanted but couldn’t find: a photo storage service that was end-to-end encrypted, fully open source, and didn’t rely on trusting the provider. “Ente” means “mine” in Malayalam, his native language.

The company now runs three products: Ente Photos (an encrypted alternative to Google Photos and iCloud), Ente Auth (a 2FA app that replaced the now-deprecated Authy for many users), and Ente Locker (encrypted document storage). The entire codebase is open source, hosted at ente-io/ente on GitHub with 23,700 stars.

Security credibility is a big part of their brand. CERN — yes, the particle physics lab — actually uses Ente and volunteered to sponsor a security audit. Cure53, a respected German cybersecurity firm, ran the audit with five researchers over two weeks. No data integrity issues were found, and 12 of 15 security-related findings were fixed during the audit itself. The cryptographic design was separately reviewed by Symbolic Software in France and Fallible in India.

Ensu is their fourth product and the first step into AI territory. The thesis is straightforward: if Ente already builds encrypted, open-source software that runs without trusting a server, why not apply the same philosophy to AI?

What Ensu Actually Does (and Doesn’t Do)

At its core, Ensu is a chat interface for small language models that run locally on your device. You download the app, it pulls a model, and you can start chatting — entirely offline.

Supported models: Depending on your hardware, Ensu runs either LFM 2.5 VL 1.6B (a lighter model for phones and lower-end machines) or Gemma 3 4B (for devices with more RAM). The download size ranges from 1.2 to 3.1 GB.

Platforms: iOS, Android, macOS, Windows, and Linux. There’s also an experimental web version. The app is available through the Apple App Store and Google Play Store, which matters — it removes the friction of sideloading or running terminal commands that most local LLM tools require.

Technical architecture: The core logic is written in Rust, shared across all platforms. Mobile apps use native frameworks; desktop apps use Tauri (a lightweight Electron alternative). This Rust core approach means features and fixes ship to all platforms simultaneously.

What’s coming: End-to-end encrypted sync of your chat history across devices, powered by your existing Ente account or a self-hosted server. This is the feature that could genuinely differentiate Ensu — no other local LLM app offers encrypted cross-device sync.

What’s limited right now: The context window is 4,096 tokens with a maximum output of 512 tokens. That’s quite small. For reference, ChatGPT handles 128K tokens and Claude supports up to 200K. With 512 tokens of output, Ensu will cut off mid-thought on anything longer than a short paragraph. Several users in the Privacy Guides community described this as “basically unusable” for anything beyond quick questions.

Ensu vs Ollama vs LM Studio vs Jan: Where Does It Fit?

The local LLM space is already crowded. Here’s how Ensu stacks up against the established players:

Ollama is the most popular option with over 100K GitHub stars. It’s CLI-first, developer-oriented, and follows a Docker-like workflow: pull a model, run it, hit the API. It supports hundreds of models and is the backbone of many local AI setups. If you’re comfortable with a terminal, Ollama gives you maximum flexibility.

LM Studio provides the most polished desktop GUI. You can browse models from Hugging Face, adjust parameters visually, and chat in a clean interface. It’s the go-to for people who want power without command-line complexity. LM Studio supports far more models than Ensu and allows fine-grained control over context size, temperature, and other parameters.

Jan positions itself as the privacy-focused GUI alternative. Cross-platform, open source, and designed for non-technical users. It’s the closest competitor to Ensu in terms of philosophy, though it offers significantly more model choices and a more mature feature set.

Ensu occupies a different niche: the simplest possible entry point. No model selection needed — it picks one for your hardware. No configuration. No terminal. Just install and chat. The tradeoff is clear: you get far less control and far fewer capabilities in exchange for zero setup friction.

The mobile support and app store distribution are Ensu’s strongest differentiators. Ollama, LM Studio, and Jan are desktop-only. If you want a local LLM on your phone that you can install from the App Store without configuring anything, Ensu is essentially the only option — apart from Google’s AI Edge Gallery, which takes a different approach with its focus on on-device function calling and agentic behavior.

The Hacker News Verdict: Mixed at Best

The 147-comment thread on Hacker News revealed a community split between those who see Ensu’s potential and those who think it shipped too early.

The critics were blunt. One commenter called it “a mere wrapper around small local LLM models” and claimed “anyone could’ve one-shotted this in Claude in an hour.” Others pointed out that Ente Photos still lacks features like RAW image support, questioning why the company launched a new product before finishing existing ones. Multiple users reported UI bugs — download progress indicators disappearing, crashes on older hardware, and incomplete documentation about RAM requirements.

Defenders pushed back on the framing. Several commenters argued that “local LLM options for less technical people is worth celebrating.” The core insight: tools like Ollama and LM Studio are great for developers, but they require comfort with terminals, model selection, and configuration. Ensu’s value isn’t technical novelty — it’s accessibility. Your parents could install it.

The trust factor came up repeatedly. Users who already rely on Ente Auth praised the company’s track record, citing the CERN audit and Ente’s consistent commitment to open source. The logic: if you trust Ente with your photos and 2FA codes, you can trust their AI app to actually be local and private.

The most interesting criticism was about timing. Ensu’s own blog post admits it’s “not as powerful as ChatGPT or Claude” and frames it as an “Ente Labs” project — an experiment. The roadmap items (persistent note assistant, LLM-backed launcher, long-term memory) generated more excitement than the current release. Several commenters noted they’re more interested in where Ensu is going than where it is today.

The Bigger Picture: Why Local AI Keeps Growing

Ensu arrives at a moment when the arguments for local AI are getting stronger, not weaker. OpenAI has started showing ads in ChatGPT. AI companies are training on user data. Enterprise customers are building entire procurement processes around data residency requirements. The EU AI Act adds regulatory pressure. And hardware keeps getting better — a modern phone with 8GB of RAM can run a 4B parameter model at usable speeds.

The counterargument is also strong: cloud AI is orders of magnitude more capable. GPT-4, Claude, and Gemini can handle complex reasoning, long documents, and multi-step tasks that a 1.6B or 4B parameter model simply cannot. Local models are good for drafting short text, brainstorming, and quick lookups. They’re not replacing cloud AI for serious work anytime soon.

Ensu isn’t trying to replace ChatGPT. It’s betting that there’s a meaningful audience who’d rather have a less capable AI that’s fully private than a more capable one that isn’t. The CERN audit, the open-source codebase, the zero-account design — it’s all building toward a specific user: someone who already chose Ente over Google Photos and wants the same philosophy applied to AI.

Whether that audience is big enough to sustain a product remains the open question. Ente hasn’t announced pricing for Ensu (the app is currently free), and the encrypted sync feature — arguably the killer differentiator — hasn’t shipped yet.

FAQ

Is Ensu free?
Yes. Ensu is currently completely free with no usage limits, no accounts, and no ads. It’s an Ente Labs project, meaning the team is still iterating on the product direction. Pricing for future features like encrypted sync hasn’t been announced, but the base offline chat functionality is free.

What models does Ensu support?
Ensu currently supports two models: LFM 2.5 VL 1.6B (for lower-end hardware) and Gemma 3 4B (for devices with more RAM). The app automatically selects the appropriate model based on your device. You don’t choose manually. The download is between 1.2 and 3.1 GB depending on the model.

How does Ensu compare to ChatGPT or Claude?
It doesn’t, in terms of raw capability. ChatGPT and Claude run models with hundreds of billions of parameters on massive server clusters. Ensu runs a 1.6B or 4B model on your phone or laptop. The output quality is significantly lower, responses are shorter (512 token limit), and the context window is much smaller (4,096 tokens vs 128K-200K). The advantage is complete privacy and offline access.

Can I use Ensu on my phone?
Yes. Ensu is available on both iOS (App Store) and Android (Google Play), making it one of the few local LLM apps with native mobile support and official app store distribution. This is a key differentiator from tools like Ollama and LM Studio, which are desktop-only.

Is Ensu open source?
Yes. Ensu’s code is part of the ente-io/ente monorepo on GitHub, which has 23,700 stars. The core logic is written in Rust, and contributions are welcome. The same repo houses Ente Photos, Ente Auth, and Ente Locker.

You Might Also Like

• Mcp2cli the Tool That Cuts mcp Token Costs by 99 Just hit Hacker News
• 685 Hacker News Upvotes in one day why Canirun ai Struck a Nerve With Local ai Enthusiasts
• Claude Channels Scores 375 Points on Hacker News Anthropics Play to Replace Openclaw
• George Hotz Wants to Sell you a 12000 ai Supercomputer and 221 Hacker News Comments Cant Stop Arguing About it
• Moltis a 60mb Rust Binary That Wants to be Your Entire ai Stack