I built a Multi-Agent AI mediator to stop "texting wars" in relationships.
Hey Indie Hackers!
I’ve spent my career in construction tech (epoxy floors, industrial sites), but last week I decided to apply AI engineering to something more complex: human emotions.
Inspired by Ed Donner’s AI track, I built SoulBuffer. It’s an MVP that acts as a safe buffer between partners.
How it works: It’s not just one prompt. It’s a Multi-Agent System. Depending on the mood, 4 specialized agents (Conflict, Flirt, Growth, Stoic) analyze your draft and suggest a version based on Non-Violent Communication (NVC) before you hit send.
The Tech:
Next.js 15
Supabase (Auth + RLS for total privacy)
OpenAI Agentic Workflow
I’m looking for two things:
Feedback on the concept: Is "Agentic Mediation" a viable niche for dating/wellness?
Technical advice: Best way to scale multi-agent calls without killing the latency?
Check out the demo: link
Let's discuss!
Congrats on the launch 👏
quick question — did you already set any hard limits on LLM/API spend?
I’ve seen a few indie AI apps get hit by agent loops early on.
Great catch! Honestly, since I'm in private beta, I haven't implemented strict credit limits yet, but you're 100% right about agent loops. I've focused on keeping the architecture simple (sequential calls) for now to avoid that, but hard caps on API spend and max iteration limits are definitely next on my list before opening it to more users. Any specific tools you recommend for monitoring?