Skip to content
FlowAI Logo FlowAI 文档

Workflow API

FlowAI Workflow API allows you to provide your designed workflows as API services, making them easy to integrate into your applications or systems. This guide will help you understand how to publish APIs, manage API keys, and how to call workflow APIs.

Part 1: Publishing and Managing APIs

API Publishing Process

To publish a FlowAI workflow as an API, follow these steps:

  1. Go to the workflow editing page you want to publish
  2. Complete the workflow design and ensure it functions properly
  3. Click the Publish API button in the top function bar publish API button
  4. Click the Save and Publish Current Workflow button publish API
  5. Confirm the operation in the popup window: “This action will save the current workflow and publish it as the latest version”
  6. Once published successfully, your workflow can be called via API API console

API Key Management

API keys are the credentials for calling FlowAI workflow APIs. Follow these steps to manage your API keys:

  1. Click the API Management option in the left navigation bar
  2. On the API Management page, you can:
    • View existing API keys
    • Create new API keys
    • View the list of published workflows and their status

Important Note: When using the API feature for the first time, you need to create an API key. Click the “Create New Key” button to generate a new API key. Please keep your API key secure and do not share it in public.

Part 2: API Call Guide

Basic Call Format

FlowAI workflow API adopts a REST style. The basic call format is as follows:

Terminal window
curl -X POST https://flowai.cc/v1/api/workflow/run/:workflow_ID \
-H 'X-API-KEY: <KEY>' \
-d '{"input":<input_parameters>}'

Where:

  • :workflow_ID: Your workflow ID (can be found on the API Management page)
  • <KEY>: Your API key
  • <input_parameters>: Parameter object defined by the workflow input node

Parameter Description

The main parameters for API requests include:

  1. Input Parameters: Different workflows require different input parameters based on the workflow design

    • Image processing workflow example: {"input":{"url":"http://example.com/image.jpg"}}
    • Text processing workflow example: {"input":{"text":"Text content to be processed"}}
    • Conversation workflow example: {"input":{"input":"Are you there?"}}

  2. Streaming Output Parameter: Add "stream":true to enable streaming output

    • Standard call: {"input":<input_parameters>}
    • Streaming call: {"input":<input_parameters>,"stream":true}

Return Format

FlowAI workflow API supports two return modes: standard output and streaming output.

1. Standard Output

In standard output mode, the API returns the result all at once after processing is complete:

Request Example:

Terminal window
curl -X POST https://flowai.cc/v1/api/workflow/run/:workflow_ID \
-H 'X-API-KEY: <KEY>' \
-d '{"input":{"input":"Are you there?"}}'

Response Example:

{
  "output": "Yes",
  "total_time": 9.316849339000001
}

Response field description:

  • output: The final output result of the workflow
  • total_time: Total execution time of the workflow (seconds)

2. Streaming Output

Streaming output mode is suitable for workflows containing Large Language Models (LLM), capable of returning the processing progress and results in real-time:

Request Example:

Terminal window
curl -X POST https://flowai.cc/v1/api/workflow/run/:workflow_ID \
-H 'X-API-KEY: <KEY>' \
-d '{"input":{"input":"Are you there?"},"stream":true}'

Response Example (Server-Sent Events format):

event:msg
data:{"Node":"myinput","Msg":"start","Type":"start","Data":{"input":"Are you there?"}}


event:msg
data:{"Node":"myinput","Msg":"Start node Start, type in","Type":"start_node","Data":null}


event:msg
data:{"Node":"myinput","Msg":"Finish node Start, type in","Type":"finish_node","Data":{"time_spend":0.000125251}}


event:msg
data:{"Node":"1741675999864","Msg":"Start node LLM1, type llm","Type":"start_node","Data":null}


event:stream
data:{"Content":"Yes","Node":"1741675999864","Reasoning":null}


event:stream
data:{"Content":",","Node":"1741675999864","Reasoning":null}


event:stream
data:{"Content":" I","Node":"1741675999864","Reasoning":null}


event:stream
data:{"Content":"'m","Node":"1741675999864","Reasoning":null}


event:stream
data:{"Content":" here","Node":"1741675999864","Reasoning":null}


event:stream
data:{"Content":".","Node":"1741675999864","Reasoning":null}


event:msg
data:{"Node":"1741675999864","Msg":"Finish node LLM1, type llm","Type":"finish_node","Data":{"time_spend":4.758483707}}


event:msg
data:{"Node":"1741676038250","Msg":"Start node Output 1, type out","Type":"start_node","Data":null}


event:msg
data:{"Node":"1741676038250","Msg":"Finish node Output 1, type out","Type":"finish_node","Data":{"time_spend":0.000128861}}


event:output
data:{"output":"Yes, I'm here.","total_time":4.758748069}

Part 3: Event Types and Example Applications

Streaming Output Event Types

Streaming output contains three main event types:

  1. event:msg: Node running log

    • start: Workflow starts running
    • start_node: Node starts execution
    • finish_node: Node execution completes, including time spent

  2. event:stream: LLM streaming output

    • Content: Currently generated text fragment
    • Node: The node ID that generates the content
    • Reasoning: Reasoning process (if applicable)

  3. event:output: Final output result

    • output: Complete output content
    • total_time: Total time spent

  4. event:error: Returned in case of errors

Example Application Scenarios

Through streaming output, you can build various real-time responsive applications:

  1. Intelligent Customer Service Bot: Display reply generation process in real-time, improving user experience
  2. Content Generator: Progressively display generated articles, code, or creative content
  3. Real-time Translation Tool: Translate as you input, reducing waiting time
  4. Educational Assistant Tool: Show thinking process, helping learners understand problem-solving steps

Part 4: Best Practices and FAQs

API Call Best Practices

Exception Handling

  • When encountering interface errors, first check the status code:
    • 401: Check if you copied the API key correctly
    • 429: Rate limit triggered, take a break before trying again
    • 5xx: Server issues, try again in a few minutes
  • For network fluctuations, we recommend:
    • Implement 3 retries with exponential backoff
    • Show user-friendly prompts like “Service is temporarily unavailable, trying to reconnect”
    • Categorize error logs: network issues/parameter errors/service exceptions

API Key Security Guidelines

Treat your API key like a bank card PIN:

  • Don’t hardcode keys in frontend projects, for example:

    // Bad practice ❌
    const API_KEY = "sk-123456...";
    

    // Correct approach ✅
    const API_KEY = process.env.API_KEY;

  • Recommend using a backend service to relay requests

  • Update keys periodically

Improving Response Speed

Optimization techniques for streaming conversation scenarios:

  • Use professional SSE client libraries (such as Python’s sseclient)
  • For high-frequency calls:
    • Set up local caching (cache identical parameter requests for 5 minutes)
    • Configure CDN acceleration for static resources
    • Use leaky bucket algorithm to control request rates (e.g., no more than 20 requests per second)

Parameter Validation

Our Workflow doesn’t validate content, so ensure input content is valid before making calls.

  1. Format validation:
    • Use whitelists for required fields
    • Check email/phone formats with regular expressions
    "input": {
      "email": "[email protected]",  // regex: ^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}$
      "phone": "13800138000"
    }
  2. Content filtering:
    • Escape special characters (<>&, etc.)
    • Filter sensitive words (political/advertising/abusive content)
  3. Boundary handling:
    • Truncate text length (prompt for modification if exceeding 500 characters)
    • Limit image size (recommended not to exceed 5MB)

Frequently Asked Questions

Q: How do I get the workflow ID?
A: You can find the ID of each workflow in the published workflow list on the API Management page. You can copy the ID and call code with one click.

Q: How do I handle errors in API calls?
A: The API returns standard HTTP error codes. You should take appropriate measures based on the error code. Common errors include authentication failure (401), parameter error (400), etc.

Q: Can I use other programming languages to call the API?
A: Yes, FlowAI API is a REST API based on standard HTTP protocol and can be called using any programming language that supports HTTP requests.