We built Celavii because we were priced out of every influencer marketing tool on the market. Annual contracts, $2K-$5K/month, per-seat fees — and none of them covered everything.

Our stack: Next.js + Supabase + LangGraph for a multi-agent system where 5 AI agents (Discovery, Strategy, Outreach, Research) orchestrate 70+ tools from plain language. The agents can do everything the dashboard does — lists, CRM, campaigns, network analysis — but from WhatsApp, Slack, or Discord.

The most interesting technical challenge was building the orchestrator. It's a LangGraph StateGraph with conditional routing — the orchestrator analyzes each request, selects the right agent, injects relevant skills from markdown files, and handles cross-agent handoffs when one agent needs another's tools. We use Gemini for orchestration and Claude Haiku for the agent nodes.

Curious if anyone else is building multi-agent systems with LangGraph. What patterns have worked for you?