AgentFlow Studio

AgentFlow Studio was inspired by the need for a lean, developer-first workbench that simplifies multi-agent AI orchestration without the bloat of heavy frameworks. Built with Python, FastAPI, and Streamlit, the platform provides an interactive environment to rapidly design, test, and execute asynchronous multi-model workflows in real time. We integrated Nebius Token Factory endpoints to serve NVIDIA Nemotron open-source models, delivering low-latency reasoning and code generation directly to the user interface. While managing state synchronization and streaming execution between FastAPI and Streamlit posed a challenge, leveraging Nebius drastically reduced compounding latency across agent loops. We are particularly proud of delivering a transparent, lightweight workbench that visualizes every step of the agent execution process. Moving forward, we plan to expand AgentFlow Studio with a node-based visual pipeline builder, web-search tool integrations, and persistent long-term memory stores.

Built With

• css3
• git
• gtihub
• html5
• nvidia
• python
• streamlit
• tavily