Show HN: I made an app to use local AI as daily driver

Original Article ↗ Hacker News Discussion ↗

Enthusiasm for a new macOS app that runs large language models locally is tempered by questions about value, privacy, and openness compared to existing free or open-source tools. Commenters weigh trade-offs between local and cloud AI (capability vs. data control), request features like system prompts, RAG over personal documents, speech interfaces, and multi-provider support, and note that Mac-only, closed-source distribution through the App Store limits both trust and reach. The broader theme is whether polished, paid native clients can justify themselves in an ecosystem already rich with cross‑platform, open alternatives built on llama.cpp and similar backends.

Scope and Purpose of the App

  • macOS app focused on local LLMs as a “daily driver.”
  • Supports Apple Silicon only; Intel Macs are not supported.
  • Uses llama.cpp directly, not Ollama, and can load custom GGUF models.
  • Some users see it as “Obsidian but with AI,” geared toward personal knowledge and local use rather than enterprise.

Online Models and System Prompts

  • App already supports GPT‑3.5/4 via user’s own API key; more configurability (custom base URLs/model names) is planned.
  • Several commenters request richer system prompt handling, multiple prompts per chat, and character/profiles.
  • Debate over how essential system prompts are: some say critical for output quality; others rarely use them.

Privacy, Openness, and Trust

  • Strong interest from users who value privacy and decentralization over centralized providers.
  • Some argue true privacy requires open-source frontends and models; closed-source apps (including this one) are viewed skeptically.
  • Concern that App Store distribution still requires trusting Apple and that “local” apps might exfiltrate data.
  • Calls for clear privacy assurances, local-only guarantees, and possibly non–App Store distribution.

RAG and Personal Data Integration

  • High demand for local AI that can search/summarize personal documents, bookmarks, browser history, email, and files.
  • Users mention existing RAG tools and want this app to support folders of markdown/txt/PDFs with sensible RAG.
  • Some skepticism about practicality of local personal-data AI since much data resides in cloud services gated by OAuth/API policies.

Features, UX, and Accessibility

  • Requested features: speech-to-text/text-to-speech, branching/editable conversations, keyboard shortcuts, Shortcuts/CLI integration, image generation, and better onboarding/model guidance.
  • Enthusiastic feedback on simplicity, design, and “feel” of the app; some users bought it immediately.
  • Others question what it does better than free/open UIs (e.g., ChatbotUI, Open-WebUI, LM Studio).
  • One thread details accessibility issues (unlabeled buttons, lack of live regions) and suggests specific improvements for screen readers.

Platform and Alternatives

  • No immediate plans for Windows or Linux versions; suggestions point Windows/Linux users to other local LLM UIs and CLI tools.
  • Some disappointment about platform limits and lack of trial, but several users consider the $10 price fair for a polished native-like experience.