Ensu – Ente’s Local LLM app

Original Article ↗ Hacker News Discussion ↗

Overall reception

  • Mixed to negative reaction to Ensu as released.
  • Many see it as “just another” local-LMM chat wrapper, while a minority are enthusiastic about a polished, cross‑platform, privacy‑oriented option.

Trust, astroturfing, and privacy positioning

  • Several comments express unease at people swiftly switching 2FA to Ente and praising it in similar language, calling the thread “ad‑like” or suspecting bots/paid shills.
  • Others counter that Ente is known in privacy circles, has had external audits, and offers open‑source clients and E2EE storage, so trusting them isn’t arbitrary.
  • Some remain wary, stressing that reputation, competence, and long‑term track record matter more than marketing claims or single endorsements.

Technical details and capabilities

  • App uses small, quantized models (e.g., LFM 1.6B, Qwen 3.5 2B/4B, possibly Gemma/Llama variants), typically 1.3–2.5 GB downloads, chosen based on device specs.
  • Users report coherent but clearly weaker performance than frontier models; suitable for basic chat, less so for complex reasoning or coding.
  • One user notes the system prompt heavily steers conversation toward Ente products, which some find off‑putting.

Comparisons to other tools

  • Frequently compared to Ollama, LM Studio, GPT4All, Jan, PocketPal, Off Grid, etc.
  • Key differentiator cited: native apps on Android, iOS, Mac, and PC with unified branding and sync, installable directly from app stores.
  • Critics see little novelty beyond being “a wrapper around llama.cpp/GGUFs.”

Product strategy and focus critiques

  • Multiple paying Ente Photos users complain about crashes, missing core features (e.g., RAW support), and confusing branding between Photos, Auth, Locker, and Ensu.
  • Some feel Ente is spreading itself thin like other multi‑product privacy companies, prioritizing new side products over stabilizing and completing the main photo service.

Use cases, audience, and “what’s next”

  • Supporters argue packaging local models into a simple, maintained app is valuable for non‑technical users who won’t run their own stacks.
  • Skeptics question who will accept major quality degradation vs. ChatGPT/Claude for privacy alone.
  • Several find the “What’s next” vision (persistent second‑brain note, launcher/agent with long‑term local memory) much more compelling than the current generic chat app.