Prelume: Find your AI Coach

Calm AI coaching that feels like opening a journal. Coaches for every life area, session reports, memory that builds over time. Creators sell access their way — IP stays fully protected.

Comment

Inspiration

I wanted to build something real in a high-pace, 4-day sprint — and Simon's brief resonated because of a simple idea: making powerful tools accessible to everyone. I don't come from the coaching world, but I know what it looks like when useful technology is locked behind complexity, expensive setups, or tools that only power users can navigate. AI coaching today has that exact problem. The potential is there, but the experience isn't. I wanted to see if I could close that gap — strip away the complexity and make it feel like starting a conversation, not configuring a system.

What it does

Prelume is a mobile-first AI coaching app with two sides:

For users: Pick a coach for the area of your life you want to work on — career, finances, family, creativity. Have a calm conversation. The app writes session reports with summaries, takeaways, and action items. It remembers what you've discussed across sessions, so each conversation builds on the last. An "About Me" profile lets you share context once instead of repeating yourself.

For creators: Build coaches with hidden instructions, generate redemption codes, and sell access through your own channels (Gumroad, website, YouTube links). Your IP never leaves the server — instructions are fully server-side and invisible to users. You keep 100% of your revenue. The app handles subscriptions; you own the customer relationship.

Users subscribe to Prelume+ for access. Creators distribute codes externally. The app never sits in the payment flow.

How I built it

Native iOS with Swift and SwiftUI. Portrait-only, minimal UI, warm linen-and-terracotta palette designed to feel calm.

Supabase handles auth (Sign in with Apple), data persistence, and five Edge Functions: streaming chat (SSE), session reports, session memory (delta tracking + global user summaries), usage credit verification, and auto-memory updates after each message.

RevenueCat powers the subscription layer — free trial, paywall gating on chat send, subscription management, and consumable token packs for heavy users hitting the weekly budget.

OpenAI runs the coaching conversations through Edge Functions, so API keys and coach instructions never touch the client.

Challenges I ran into

Building a full product in 4 days with zero hands-on experience in Supabase, OpenAI, or RevenueCat paywalls. Every integration was a first. RevenueCat deserves a shoutout here — their SDK made monetization the easiest part of the stack to get running.

Session memory across conversations was a real design challenge. I needed the AI to remember context without overwhelming the prompt or blowing token budgets. The solution — delta tracking per message plus compressed global user summaries — keeps memory useful and lean.

Accomplishments that I'm proud of

  • Feature depth in 4 days. Streaming chat, conversation history, session reports, session memory, discover marketplace, About Me profiles, usage limits with token packs, an orchestrator agent with tool calling, deep linking, home screen widgets, Siri shortcuts, Spotlight indexing, markdown rendering, toast notifications, and a creator prompt editing flow — all native, all working, solo.

  • 30-second time to first conversation. Open the app, sign in with Apple, pick a coach, start talking.

  • Session reports that actually help. Automatic summaries, takeaways, and action items in a coach-to-client tone — not a transcript dump.

  • Memory that compounds. The app genuinely gets better the more you use it. Session 5 feels different from session 1.

  • Creator IP model. A distribution system where creators keep full control and full revenue.

What I learned

Design is the product, not the features. Simon's brief emphasized calm, minimal, and human-centered — and every time I prioritized polish over adding another feature, the app got better. The warm color palette, the absence of clutter, the immediate path to a conversation — these matter more than the feature count.

Creator-side thinking changes everything. Building for two personas (creator + user) forced me to think about business models, not just UX. The redemption code model — where the app never handles creator payments — is simpler AND better for creators than any marketplace cut.

What's next for Prelume: Find your AI Coach

  • Creator dashboard & API — analytics, code management, and an API for creators to integrate coach distribution into their own platforms and funnels.

  • Contextual file sharing — let users bring documents, images, and links into coaching sessions for richer conversations.

  • Deep data access — coaches that can tap into Apple Health, Calendar, Screen Time, and other on-device data to give truly personalized guidance.

  • Refined coach creation flow — making it faster and more intuitive for creators to build, test, and iterate on coaching instructions.

  • Launch with Simon — if we move forward together, I'd adapt coaching instructions specifically for the Better Creating philosophy, optimized for mobile-first everyday users.

Built With

  • openai
  • revenuecat
  • supabase
  • swift
Share this project:

Updates