AI-Flow is the open source AI workflow builder for composing multi-model pipelines through a visual canvas — connecting LLMs, image generators, APIs, and file inputs into automated workflows without writing integration code.
The Problem
n8n and Zapier handle general automation well but were not designed for AI model composition. Chaining GPT-4 for summarization, DALL-E for image generation, and a custom storage API requires custom nodes, awkward data mapping, and no native handling of prompt templates or model parameters. Proprietary AI workflow platforms like Relevance AI and Dust are cloud-only, charge per execution, and cannot run behind your own firewall.
How AI-Flow Solves It
AI-Flow provides a node-based canvas where each node is an AI model, a prompt template, or a data operation. Connect nodes to define the flow of data: route the output of one model into the prompt of the next, branch on classification results, loop over a list of inputs, and write results to a file or API endpoint. Supports OpenAI, Anthropic, Stability AI, Replicate, and custom OpenAI-compatible endpoints.
Key Features
- Visual drag-and-drop canvas for building multi-model AI pipelines
- Prompt template nodes with variable injection from upstream model outputs
- Supports OpenAI, Anthropic, Stability AI, Replicate, and custom API endpoints
- Image generation, text completion, classification, and embedding nodes built in
- Batch execution: run a workflow over a list of inputs in one operation
- Export workflows as JSON for version control and team sharing
Who It's For
AI-Flow is best for developers prototyping multi-model AI pipelines before committing to a custom implementation, content teams automating workflows that involve several AI services, and organizations that want visual AI automation running on their own infrastructure without a cloud AI workflow subscription.
Compared to n8n or Zapier
Unlike n8n or Zapier, AI-Flow is purpose-built for AI model composition — prompt template nodes, model parameter controls, and native handling of LLM inputs and outputs are built in rather than bolted on through generic automation nodes.
