Build, export, and run multi-agent workflows with the OpenAI Agents SDK

GPT builders
, ,
1

We’ve shipped a small toolchain around the Agents SDK:

  • Dashboard: https://dashboard.multisync.io
    Design agents, pick output schemas (we enforce a required result), attach MCP servers, and assemble flows (single agent or agent-reviewer).

  • Export: Download a self-contained agents-compose.json.

  • Run locally: npm i -g multisync → multisync --config agents-compose.json (uses your OPENAI_API_KEY).

Features

  • Agents-as-tools (serial calls)

  • Reviewer loops (LLM-as-judge) with pass conditions (result.score >= …), feedback injection mode, and max turns

  • JSON Schema / structured outputs validation

  • MCP stdio/HTTP integration

  • Verbose, colorized logs

Repo: GitHub - Multi-Sync/multisync
Site: https://multisync.io

Would love feedback from the community :hugs: