Problem & Context
Relationship coaching and therapy are expensive, often unavailable when you need them most, and limited to scheduled sessions. When you're in the middle of a difficult conversation or processing something at 11pm, there's nowhere to turn.
So people are turning to AI. My partner and I started doing this too. It helped, but something felt off. The AI would agree with whoever was talking to it. If I vented about a disagreement, it took my side. When she did the same, it took hers. We weren't getting clearer, we were getting more entrenched.
The key opportunity: What if the AI was honest instead of agreeable? Same rules for everyone. Same willingness to challenge your framing. Same question: "How would your partner describe this situation?"
Approach
SAM is an AI relationship coach you can text. No app to download, no new interface to learn. Save the number, start a conversation. It meets you in the messaging environment you already use every day.
Because this is sensitive territory, every design decision prioritizes trust:
- Privacy by default. Conversations are encrypted with your own passphrase. I literally cannot read what you discuss with SAM, even if I wanted to.
- No weaponization. SAM refuses to help draft manipulative messages, guilt trips, or ultimatums. It helps you communicate better, not win arguments.
- Safety awareness. SAM watches for signs of self-harm or crisis. When conversations go beyond coach-level support, it exits gracefully with resources, without penalizing the user or making them feel broken.
- Value before monetization. The goal is delivering a genuine "aha moment" before ever asking for anything. If SAM isn't useful, it shouldn't cost you.
Tradeoffs:
- Telegram first. Started here because it's the fastest to build and test. WhatsApp and SMS will come after validation.
- Honesty over helpfulness. SAM pushes back and challenges your perspective. This can feel less "supportive" in the moment, but it's more useful.
- Privacy over debugging. User-keyed encryption means I can't access conversations to improve the product. Worth it for trust.
North Star Metrics
Primary: First value. Did the user have an "aha moment"? A long conversation isn't bad; it often signals deep engagement. The hurdle is whether SAM delivered something genuinely useful.
Secondary: Continued engagement. A mix of frequency (do they come back?) and depth (do they open up?). If someone returns and goes deeper, that's the signal.
Tech Stack
- TypeScript / Node.js
- Telegram Bot API (grammy)
- Anthropic Claude API for the coaching intelligence
- PostgreSQL for storage, Redis for sessions
- Deployed on Railway.app
Status
Currently in development. Core features are built: coaching personality, conversational onboarding, encryption, and the safety system. Now testing and running AI evals to refine the experience before opening up more broadly.
Want early access?
SAM is currently in development. Join the beta waitlist to be first in line when it's ready.