AIFLOW logo

AIFLOW

Open standard for multi-agent workflows

v0.1.0 — Open Standard Preview

The open standard for multi-agent workflows.

AIFLOW lets you define, share and execute AI agent workflows using a single portable .aiflow file. Design visually in AIFLOW Studio, run them via browser or CLI runtimes, and keep everything in one spec.

View on GitHub → Download v0.1.0 View .aiflow spec Explore examples
React-based Studio • Node CLI runtime • Single JSON .aiflow spec

Quick start 

# Clone and launch the AIFLOW Studio
git clone https://github.com/AIFlow-studio/AIflow
cd AIflow
npm install
npm run dev

# Run a workflow via the CLI runtime
export API_KEY=YOUR_GEMINI_API_KEY
npm run run-flow -- ./examples/CustomerSupportFlow/

Why AIFLOW?

Open standard

One portable workflow format

AIFLOW describes agents, models, routing logic and memory inside a single .aiflow JSON file. Easy to version, share and review in Git.

Studio + runtime

Design and execute cleanly

Design multi-agent workflows visually in AIFLOW Studio, then execute them via the browser or Node CLI runtime. No hidden glue code, fully declarative.

Multi-provider

Choose your AI stack

Mix OpenAI, Gemini, Anthropic, Groq and more inside one workflow. Each agent defines its own provider, model, role and prompt.

Example workflows

Browse all examples →

Customer support

CustomerSupportFlow_v1.0.0.aiflow

Multi-agent ticket triage with classifier, engineer and responder steps. Ideal for L1 support automation.

npm run run-flow -- ./examples/CustomerSupportFlow_v1.0.0.aiflow

Marketing

MarketingContentFlow_v0.5.0.aiflow

Go from content idea to strategy, long-form article and social posts, all inside one automated workflow.

npm run run-flow -- ./examples/MarketingContentFlow_v0.5.0.aiflow

Sales

LeadQualificationFlow_v1.0.0.aiflow

Lead intake, qualification and sales briefing. Perfect for B2B funnels and SDR teams.

npm run run-flow -- ./examples/LeadQualificationFlow_v1.0.0.aiflow

AIFLOW specification v0.1

View spec on GitHub

Structure

One JSON file, full workflow

AIFLOW describes metadata, agents, routing logic, tools and memory in a single portable JSON document. 

{
  "metadata": { ... },
  "flow": { ... },
  "agents": { ... },
  "tools": { ... },
  "memory": { ... }
}

Execution model

Deterministic multi-agent routing

The runtime executes agents step by step, parses model output into JSON and uses declarative conditions to determine the next agent.

  • Start at flow.start
  • Run agent prompt + AI model
  • Parse JSON output into context
  • Evaluate conditions in flow.logic
  • Stop when no transitions remain

Roadmap

v0.2

  • Expression-based routing
  • Improved debug console
  • Better CLI UX

v0.3

  • Tooling standard
  • Built-in HTTP & DB tools

v0.4

  • Shared core runtime
  • Python runtime

v1.0

  • Stable AIFLOW spec
  • Validator & schema
  • Plugin ecosystem