Loomflow

Draw your agents

The large view above is a multi-agent supervisor flow: a supervisor model routes work to a software engineer and a code reviewer, loops back until the job is done, then writes the final answer.

The two views below show the Agents page, where every saved flow lives, and an agentic RAG flow that checks whether retrieved documents are relevant before answering.

Chat flows and RAG

Studio assembles chat flows from models, prompts, retrievers, vector stores and document loaders, so a knowledge-base assistant takes minutes, not days.

Document stores handle chunking, embeddings and vector upserts, and ready-made templates give you a working starting point for the common patterns.

One place for resources

Tools, credentials, variables and API keys live together, so every flow reuses the same connections instead of copying secrets around.

Runs record every execution so you can inspect what each agent did and judge whether the answer was any good.

Under the hood

Loomflow has a React interface over a Node.js and TypeScript server, and any flow can be exposed as an API or an embeddable chat widget.

The templates screen below lists ready-made starting points for common patterns, using LangChain nodes for models, memory and retrieval.

Open source

The full code and a two minute walkthrough video are on GitHub, along with screenshots of every screen.

Highlights

A warm coral interface, a canvas with colour-coded nodes, a menu grouped into Agents, Studio and Resources, and demo flows with docs and a walkthrough video.