Run AI agents as a workflow
Chain agents on a canvas. Each gets its own task, model, and tools, and passes its result to the next. Read every conversation, and step in at any point.
Skip to content
- Draw the workflow
- Each step is an agent with a prompt, a model, and tools. Connect steps to say which results feed which.
- Run steps in parallel
- Steps that don't depend on each other run at the same time. A step starts when everything it needs is ready.
- Stay in control
- Open any agent's conversation to answer its questions, approve a tool call, or retry a failed step.
- Bring your own model
- Use OpenAI, Anthropic, ChatGPT sign-in, Amazon Bedrock, OpenRouter, or a local model such as Ollama.
How a run goes
Draw the workflow. Add an agent for each step. Give it a prompt, a model, and tools, then connect it to the steps that feed it.
Run it and watch. Steps that don't depend on each other run at the same time. Open any agent's conversation to answer a question, approve a tool call, or retry a failed step.
Review what happened. History keeps every run. Open a replay to read each agent's conversation and the handoffs between them.
Get started
- Install OpenFlow. The macOS app needs no development tools.
- Connect a provider so agents can call a model.
- Build your first workflow, or start a direct chat if you only want to talk to one agent.
Find what you need
| I want to… | Read |
|---|---|
| Understand how it works | How OpenFlow works |
| Design a workflow | Workflow authoring |
| Handle approvals, retries, and failures | Runs and recovery |
| Look up a setting | Configuration |
| Fix a problem | Troubleshooting |
| Build it from source | Build from source |
| Browse everything | All documentation |