In this quick episode, we share our newest feature—a privacy‑first ✏️ spell checker for your private journal—and walk through how we’ve built AI into MOSSLET without turning your thoughts into training data. From journaling prompts and handwriting OCR to mood insights and image safety checks, we explain exactly what gets sent, where it goes, and how we keep your words yours. 🌱
Extended Show Notes (full description):
In this episode, we pull back the curtain on how we’re building AI that helps without harvesting inside MOSSLET, our small, bootstrapped, privacy‑first social network with a private journal at its core. 🌿
We cover:
- Our new spell checker ✏️ for your private journal, and why we designed it to help you write better without profiling you
- How journaling prompts work with just a single mood word 😊 instead of your full journal history
- What happens when you upload a handwritten journal page 📷: image → text → immediate deletion
- How mood insights use dates, mood labels, and word counts—never your raw entries 💭
- Our three‑layer privacy approach: send only what’s necessary, process and delete, and contractual protection with data_collection: "deny"
- How we use local AI models ⚙️ (via Bumblebee in Elixir) as a fallback so many requests never leave our own servers
- Why we haven’t moved AI fully into the browser yet—and what would need to change for that to make sense 💻
If you’ve ever wondered, “How can a journaling app use AI and still be private?”, this episode walks through the real technical and product decisions behind our answer.
Have questions or want us to dive deeper into the technical side? Reach out—we’re building this in public and are happy to talk about how it all works under the hood.
Episode 4
