If you're like me, you're probably using multiple AI tools daily: - Claude for complex reasoning and code - Cursor/Copilot for IDE assistance - Aider, Claude Code, or similar coding agents - n8n/Make for automations - Custom agents via API Here's the problem: we're not using these tools to their full potential. Why? Because we don't fully trust them to act autonomously. So we: - Watch every action they take - Manually copy-paste instead of letting them execute - Avoid "agentic" features that could go wrong - Keep them on a short leash - The result? We're babysitting AI instead of leveraging it. The missing piece: Human approval at critical moments. I built ViaHuman to solve this. It's a simple concept: Your AI agent reaches a critical decision point It pauses and sends a push notification to your phone You approve, reject, or modify - with one tap The agent continues Now I can: - Let Claude Code run autonomously → approve before it commits/deploys - Let my email agent draft responses → quick review before sending - Run n8n workflows at 2am → get notified only when human judgment is needed - Actually use "agentic" mode without anxiety For developers using multiple LLMs: - MCP Server - works with Claude Desktop, integrates with any MCP-compatible tool - Mobile app - approve from anywhere, don't be chained to your desk - Context-rich requests - see what the agent is asking, with full context

ViaHuman

Human in loop notifications for the LLMs

Human in loop notifications for the LLMs