Launch HN: Martin (YC S23) – Using LLMs to Make a Better Siri
A new YC-backed iOS app called Martin aims to be a smarter, more proactive personal assistant than Siri by using large language models to manage reminders, calendars, texts, and (eventually) email. Commenters are intrigued by its ability to act on users’ behalf—like scheduling meetings or texting contacts—but raise strong concerns about privacy, data access, reliability, and the $30/month pricing, especially with Apple’s own LLM-powered Siri upgrade imminent. Many see potential in third‑party assistants that integrate deeply with personal data, but question whether users will trust small startups over platform vendors and whether such products can stay viable once native OS features catch up.
Product concept & capabilities
- iOS app aiming to be a “better Siri”: voice-first personal assistant that integrates with calendar, reminders, email, SMS, and (planned) docs.
- Can text contacts from its own number and will auto-continue the conversation; opens messages with a clear indication it is an assistant.
- Core use cases: daily schedule syncs, reminders, meeting planning, dictation/transcription, brainstorming, and task capture.
- Uses GPT and Claude under the hood, plus RAG and reflection-based “memory” over time.
User experience reports
- Some users are impressed: it correctly handles fairly complex multi-step scheduling tasks in one shot and feels more useful than generic chatbots.
- Others report serious reliability issues: app crashes, SMS not responding, emails not actually sent despite confirmations, weak web research, hallucinated company descriptions, and long delays or no responses.
- Onboarding is criticized as confusing, especially around required calendar connection and unclear current limitations (e.g., email sending not yet live).
Integrations, roadmap & technical approach
- Most-used integrations: calendar and reminders; morning sync is common.
- Team targets one major new integration per month; Outlook/Exchange and document editing via Google/Word are on the roadmap.
- Long‑term memory: combination of embeddings plus periodic LLM “reflection” at conversation, daily, and multi‑day goal levels.
- Users strongly request “bring your own LLM/API key” and a clearer integrations list.
Pricing, trial, and business model
- $30/month subscription viewed by some as steep for an early-stage product; others note token costs and development effort justify it.
- 7‑day trial feels too short to many for habit‑changing workflows; credit-card requirement is a deterrent.
Competition with Apple/Google & “Sherlocking” risk
- Large debate over whether Apple’s upcoming “Apple Intelligence” and LLM-powered Siri will effectively obsolete such products.
- Some argue Apple will win via distribution and deep on-device context; others think Apple moves slowly, leaves many niches, and not all users fit Apple’s rigid patterns.
- Several commenters question YC funding so many assistants that are “thin wrappers” over third‑party LLMs.
Privacy, security & trust
- Strong concerns about granting deep access to email, calendar, messages, and calls.
- Product cites CASA Tier‑2 and Google OAuth reviews; SOC 2 is “planned.”
- Users ask for explicit answers on: data sent to OpenAI/Anthropic, training usage, deletion rights, encryption practices, and clear guarantees against data sale or ad targeting.
- Some plan to wait for Apple’s solution, perceiving its privacy posture as stronger; others distrust all large providers equally.