Overview
What is Kiwi Flow?
What Kiwi Flow is, how workflows are built and run, and how it relates to Kiwi Fabric and Kiwi Distro.
Kiwi Flow is a workflow automation and orchestration platform. You model a business or technical process as a workflow, a graph of activities, and the platform runs it, records every step, and lets you evolve it over time.
Workflows can be long-running (hours or days), run branches in parallel, pause for human approval, handle errors explicitly, and call out to cloud services, databases, APIs, remote machines and desktop applications.
The pieces#
| Component | What it does |
|---|---|
| Workflow Studio | A browser-based visual designer with a drag-and-drop canvas, inline scripting and live validation. It produces workflow JSON. |
| Flow engine | Loads a workflow definition, executes its activities, resolves inputs and emits feedback for every step. |
| Backend API | REST API for workflows, runs, triggers, connections, administration and AI generation. |
| Admin area | Users, roles, groups, audit log, API keys, resource guards and licensing. |
| Client SDKs and CLI | Python, .NET, Java and Node.js SDKs, and the kiwi flow command line. |
What a workflow is made of#
A workflow is a JSON document with inputs, global variables and a list of activities. Each activity has a type (for example http_request, sql_database, llm_text, if_else), maps its inputs from earlier results, and points to the next activity. See Build a workflow.
The platform ships 84 activity types across flow control, integrations, databases, cloud services, messaging, SSH, LLM, files, caching, and first-party connectors for Kiwi Fabric and Kiwi Distro. See the activity catalogue.
Ways to author#
- Draw it in Studio.
- Write or edit the JSON directly (the Studio has a code view; the CLI imports and validates files).
- Describe it and let an AI model generate the workflow, then refine it with follow-up prompts. See Generate workflows with AI.
- Any mix of the above.
AI in two places#
Kiwi Flow uses AI to build workflows (generate and enhance) and to run intelligence inside them with four activity types: llm_text, llm_decision, llm_image and llm_rag. It works with OpenAI, Azure OpenAI, Anthropic and Ollama, using your own provider keys. See Use LLM activities.
Reusable building blocks#
A workflow marked reusable can be embedded in other workflows as a sub_workflow. Common logic such as identity checks, notifications or approvals can live in one place and be versioned and governed independently.
Together with Kiwi Fabric and Kiwi Distro#
Each Kiwify Labs product stands alone, and they combine well:
- Kiwi Fabric gives workflow steps federated SQL and distributed transactions across many data sources without embedding database coupling in the workflow. See Query and write data with Kiwi Fabric.
- Kiwi Distro runs registered scripts (AutoIT, Playwright, Python, Node, PowerShell, batch, shell, AutoHotkey) and method calls on remote workers, for example thick-client desktop automation on a Windows machine. See Run scripts and remote work with Kiwi Distro.
Deployment#
Kiwi Flow is deployed in the cloud or on premises. Data access in the platform is database-agnostic (through the Kiwi ORM), with PostgreSQL recommended for new installations.