Pipedream is an integration platform for developers.
We provide a free, hosted platform that makes it easy to connect apps and develop, execute and maintain event-driven workflows. The platform has over 400 fully integrated applications with managed authentication and support for over 1M npm packages.
Key Features:
Product Demo: YouTube (5 minutes)
You can also get support, raise a bug or feature request, or file a security disclosure.
Workflows are a sequence of linear steps - just Node.js code - triggered by an event (via event source, HTTP endpoint, timer, and more). Workflows make it easy to transform data and integrate with 300+ APIs from various apps and services.
See our workflow quickstart for a detailed walkthrough of building a workflow end-to-end.
As you build more advanced workflows, you may also find these docs helpful:
Pipedream receives data via event sources. Event sources are open source, run on Pipedream's infrastructure and collect data from your own application and/or services like GitHub, DropBox, Zoom, RSS feeds, and more.
Event sources emit new events produced by the service, which can trigger Pipedream workflows, or which you can consume using Pipedream's REST API or a private, real-time SSE stream.
Here is the simplest event source, an HTTP source:
module.exports = {
name: "http",
version: "0.0.1",
props: {
http: "$.interface.http",
},
run(event) {
console.log(event); // event contains the method, payload, etc.
},
};
Popular Event Sources:
Event sources can also be deployed via the Pipedream CLI. Once installed, you can deploy an event source by running:
pd deploy # prompts you to select a component and pass required options
You can create your own event sources for your own personal use. If you think others would benefit from your source, you can publish them to all Pipedream users. See these docs to get started:
Actions are prebuilt code steps that you can use to perform common operations across Pipedream's 400+ API integrations, for example: sending email, adding a row to a Google Sheet, and more. Pipedream supports thousands of prebuilt actions.
Typically, integrating with these services requires a custom code to manage authentication, error handling, etc. Actions abstract that for you: you just pass the necessary params as input, and the action handles the rest. For example, the HTTP GET Request action accepts the data you want to send and the URL you want to send it to, returning the HTTP response for use in future steps.
You can create your own actions, allowing you to re-use them across workflows in your account. You can also publish actions to the entire Pipedream community, making them available for anyone to use.
Here's the code for an action to make an HTTP GET request:
const axios = require('axios')
const http = require('../../http.app.js')
module.exports = {
key: "http-get-request",
name: "GET Request",
description: "Make an HTTP `GET` request to any URL. Optionally configure query string parameters, headers and basic auth.",
type: "action",
version: "0.0.1",
props: {
http,
url: { propDefinition: [http, "url"] },
params: { propDefinition: [http, "params"] },
headers: { propDefinition: [http, "headers"] },
auth: { propDefinition: [http, "auth"] },
},
methods: {},
async run() {
const config = {
url: this.url,
method: "GET",
params: this.params,
headers: this.headers,
}
if (this.auth) config.auth = this.http.parseAuth(this.auth)
return (await axios(config)).data
},
}
Destinations, like actions, abstract the connection, batching, and delivery logic required to send events to services like Amazon S3, or targets like HTTP and email.
For example, sending data to an Amazon S3 bucket is as simple as calling $send.s3()
:
$send.s3({
bucket: "your-bucket-here",
prefix: "your-prefix/",
payload: event.body,
});
Pipedream supports the following destinations today:
Pipedream has a generous free tier. You can run sources and workflows for free within the limits of the free tier. If you hit these limits, you can upgrade to one of our paid tiers.
The Pipedream platform imposes some runtime limits on sources and workflows. Read more about those in our docs.
Before adding an issue, please search the existing issues or reach out to our team to see if a similar request already exists.
If an issue exists, please add a reaction or comment on your specific use case.
If an issue doesn't yet exist, please use these templates to create one:
You can read about our platform security and privacy here.
If you'd like to report a suspected vulnerability or security issue, or have any questions about the security of the product, please contact our security team at security@pipedream.com.
数据科学是艰苦的工作,而不是魔咒。 一个AI模型的性能是否如所宣传的取决于其训练的程度,并且没有“一刀切”的方法来训练AI模型。 分布式AI培训的必要弊端 缩放是训练AI模型时最棘手的考虑之一。 当模型变得过于资源匮乏而无法在任何单个计算平台上进行整体处理时,培训将尤其具有挑战性。 模型可能增长得很大,超过了单个处理平台的内存限制,或者加速器需要开发特殊的算法或基础结构。 训练数据集可能会变得如此