TL;DR: Mobile notifications for AI workflow approvals. Built it because running 5+ AI agents at once is chaos.
The problem:
This paragraph is the longest thing I've written by hand in weeks. Everything else? AI-generated across 5 browser tabs (GPT, Gemini, Claude) and multiple terminals running Claude Code.
They're FAST. They're GOOD. But here's what nobody tells you: you become the bottleneck.
I'd start 3 agents on different tasks, context switch between them, and lose track of which ones needed my input vs which ones were safe to auto-execute. Critical decisions got buried in terminal output.
What I tried:
Telegram bots (hacky, unreliable)
Slack integrations (too much noise)
Checking terminals every 5 minutes (defeats the purpose of automation)
None of it worked.
What I built:
viahuman – a notification server that lets AI agents ping me when they need human approval.
How it works:
Integrated a local MCP server with my LLMs
Told them: "Run autonomously, but notify me for important decisions"
They send mobile push notifications when done or stuck
I approve/provide context from my phone
They adapt and continue
Telegram bots (hacky, unreliable) Slack integrations (too much noise) Checking terminals every 5 minutes (defeats the purpose of automation)
None of it worked. What I built: viahuman – a notification server that lets AI agents ping me when they need human approval. How it works:
Integrated a local MCP server with my LLMs Told them: "Run autonomously, but notify me for important decisions" They send mobile push notifications when done or stuck I approve/provide context from my phone They adapt and continue