We wanted to create a AI orchestrator that manages various workflows to produce a trusted answer and provides services like Polymarket event updates, creating scripts for TikTok videos, and summarizing emails. You ask about crypto, it should hit Polymarket. You ask for a TikTok script, it should run a video-script workflow. You ask about the weather or your email, it should talk to the right service. That’s what inspired us to build Agent Hive – a small “brain” that coordinates multiple specialized agents and workflows behind one chat interface.

We built the system around n8n as our orchestration layer. We also used Eleven Labs templates for workflows and Polymarket's API to get updates on events. On top of that we wrote a Task Manager / Orchestrator in TypeScript: it takes the user’s natural-language request, plans which workflow(s) to call, sends them structured JSON, then aggregates all the responses and passes a clean context JSON to the LLM (Gemini via OpenRouter) to generate the final answer.

We learned a lot about deploying orchestrator through Railway and its integration: -Designing JSON files so that the orchestrator could understand the workflows and pick and combine them. -Using n8n webhooks so that the data could be passed between workflows -Handling big, noisy API payloads (like the Polymarket events you saw) and turning them into something an LLM can safely reason over

We hit a lot of walls, honestly: deploying, creating clean JSON files, combining workflows, and the orchestrator as a whole. We used Cursor and Eleven Labs templates for reference to understand the architecture.

Built With

Share this project:

Updates