TypeScript icon, indicating that this package has built-in type declarations

3.0.1 • Public • Published


@pieces.app/pieces-os-client is an open source on-device AI development workflow assistant.

This package contains the endpoints for communicating with Pieces OS to add Pieces Copilot conversations or save code snippets and resources - entirely offline and on device with Local Large Language Models (LLLMs) keeping your data secure.

Follow this guide to get started with the Pieces Client in your own development environment. Check out the table of contents below to understand specific functionality, but we recommend reading along the Copilot Series

npm version npm version npm version

Table of Contents


Operating System Support

  • macOS
  • Windows
  • Linux (see specific distributions here)

Configuration & Setup with NPM

You can choose to follow the following steps to configure your project by hand, or you can use npx create-react-app <your-app-name> instead.


When developing on the Pieces platform, you need two primary things:

  1. Download the Pieces OS application
  2. Install the npm package

Pieces OS

Pieces OS runs in the background of your computer and serves as a hub for all plugins and extensions developed by the team. In order to utilize your own Server locally and support all the functionality that powers things like Global Search, Copilot Chats, Asset Saving, context, and more.

Select the right version to download Pieces OS for your operating system:

  • macOS - [Compatible with macOS 11 Big Sur or higher]
  • Windows - [Compatible with Windows 10 version 1809 or higher]
  • Linux - [Compatible with Ubuntu 18 or Higher]

You can also visit our user facing documentation to learn more about different features that are available now to give you an idea of some of the things that you can potentially do.

Installing Node using NVM

To install the current version of node, simply run the command:

nvm install --lts

To check what version of node that's currently running on your machine, run the command:

node --version

To install a specific version of node, you can run the command:

nvm install [version-of-node]

To switch between different versions of node, you can run the command:

nvm use [version-of-node]

npm comes installed when you use nvm to install node.

Downloading NPM Package

Using npm:

npm install @pieces.app/pieces-os-client

Using pnpm:

pnpm add @pieces.app/pieces-os-client

After you install the package, you can import the library into your file(s) using require:

const pieces = require('@pieces.app/pieces-os-client')

or you can import the package using import as well:

import * as pieces from '@pieces.app/pieces-os-client'

Recommendation The order that npm packages are saved and added to your dependencies is important and will affect your installation flow. This slowed me down for quite a bit.

If you are having issues with your installation, it is likely due to a conflict in Typescript versions - npm uninstall typescript - then go back and perform all other npm installations before reinstalling typescript again.

You can get it here: GitHub Repo

Creating a Project from Scratch

Let's get started with the base of your new React project! We will do this by learning more about manipulating Pieces OS and creating our own assets locally on device. First, we will create a directory for you project along with some dependancy setup:

  1. Create a directory for your project called (or the name of your choice)
  2. Open up a Terminal at the root of that directory, and run npm uninstall typescript to ensure you are starting fresh
  • (optional) run nvm install --lts to check your version of Node
  1. From that same terminal window, use the prefix parameter in your installation command to install your dependencies: npm install --prefix ./ @pieces.app/pieces-os-copilot

  2. Install Dependencies with NPM with these three commands:

npm install typescript && npm install ts-node

# Useful types package
npm install -D tslib @types/node

5 (optional) additional npm packages

npm install @types/react && npm install @types/react-dom

npm install react && npm install react-dom && npm install react-scripts
  1. Add these scripts into your package.json to help with development, and be sure to save after adding them:
  "scripts": {
    "dev": "ts-node index.tsx",
    "clean": "rm -r node_modules && rm package-lock.json",
    "start": "react-scripts start",
    "build": "react-scripts build"

Don't Forget You can run any script that you add here via npm run <your-script>. For example to run "start" you would simple run npm run start

Setting up your public Directory

Next you can go ahead and create a new directory called public that will hold your index.html file where your entry point exists. Create the file inside of public and save it there. You do not have to add anything to the index.html file at this time, as we will come back to this later. If you would like, you can add the following as a placeholder for now:

<!DOCTYPE html>  
<html lang="en">  


Setting up your src Directory

Now that the initial .html file has been created, you can start to work on your src directory and getting the rest of your core files added to the project.

Inside the src directory, add two files:

  • singleton.ts - single instance container for a few global functions and features
  • index.tsx- where the core info is and where we are going to be spending most of the time during this project following the setup.

Once you open index.tsx you should follow these steps to get your base Application window created:

  1. Import the full react library at the top of your file, along with a single import from react-dom, and the createRoot method:

    import * as React from "react";
    import { createRoot } from "react-dom/client";
    import { render } from 'react-dom';
  2. Follow that with the full App() function and main run of the application:

    function App(): React.JSX.Element {  
        return (  
                <h1>Hello Pieces Dev Community :)</h1>  
  3. And then add these last two lines to target the root element by using the ID that is on it:

    const rootElement = document.getElementById("root") as Element;  
    const root = createRoot(rootElement);

Helpful tip When working in this environment I noticed some @babel errors during the build process (or running npm run start) and found this workaround that you can install via npm as well here.

Running your Project for the First Time

Everything has been added. We are nearly there and will need to perform a final few checks before starting our dev project.

  1. Be sure that Pieces OS is running
  2. (optional) Run another npm install (because it never hurts)

Now that everything is ready to go, lets run this command here:

# remember that we added this to the scripts as: "start": "react-scripts start"
npm run start

And after a few seconds you should be able to see in your Chrome browser (or your primary browser) a blank window that looks like this:

You have now successfully set up your dev environment, and will be ready to test different endpoints inside of Pieces OS.

Connecting your Application

When Pieces OS is running in the background of your machine, it is communicating with other local applications that use Pieces software, and up until recently only supporting internally built tools.

As each plugin, extension, or application initializes they 'reach out' to Pieces OS and authenticate with the application themselves. There are a number of application formats that we support and provide for each of our applications. When developing on Pieces OS, you can use "LOCAL_DEV" to avoid any issues with other applications.

Creating Application

The application model describes what application a format or analytics event originated from. This is passed along when initializing your dev environment and creates a connection to Pieces OS.

To create the Application object for your project, you will need to make sure that you have the following two things:

  1. Create a tracked_application json object to hold your formatted data
  2. Output using console.log() following your connect() method is complete

We will go over the different usages in each of these steps one at a time to bring up any specificities that may need a deeper explanation.

Using tracked_application

Connecting your application here is as easy as a single POST request and can be done via the Response interface of the Fetch Api. Remember that you can name this whatever you would like to, just be sure to include the updated variable name in the options down below.

When creating the tracked_application item, you will need to use a type that is not available inside the current npm_deployment.

This structure is the same as the tracked_application full example you see here below, and the only difference from the unavailable type SeededTrackedApplication and the available type TrackedApplication is id: number.

First lets take a look at the tracked_application object:

const tracked_application = {  
    name: Pieces.ApplicationNameEnum.OpenSource,  
    version: '0.0.1',  
    platform: Pieces.PlatformEnum.Macos,  
  • name: OpenSource
  • platform: Depending on your current environment, you need to set the platform parameter to match your current operating system. Select between .Macos, .Windows, .Linux
  • version: A string value that can be any string value and can be used to track the application you are building

Imports Be sure to double-check that you have the following import added to the first few lines of your index.tsx file if you have not already: import * as Pieces from "@pieces.app/pieces-os-client";

Connect using Pieces.ConnectorApi() and tracked_application

When your program starts, it needs to connect to Pieces OS to gain access to any functional data and to exchange information on the localhost:1000 route. Now that you have your tracked_application - lets get into the details.

Start by defining you connect function and prepare the connectorApi on Pieces.ConnectorApi().connect() passing in the tracked_applicaition we created above:

export async function connect(): Promise<JSON> {
  const connectorApi = new Pieces.ConnectorApi();
  const response = await connectorApi.connect({
    seededConnectorConnection: { application: tracked_application },
  return JSON.parse(JSON.stringify(response));

Here is the entire connect function for you to double-check your work:

const tracked_application = {
  name: Pieces.ApplicationNameEnum.OpenSource,
  version: '0.0.1',
  platform: Pieces.PlatformEnum.Macos,

export async function connect(): Promise<JSON> {
  const connectorApi = new Pieces.ConnectorApi();

  const response = await connectorApi.connect({
    seededConnectorConnection: { application: tracked_application },

  return JSON.parse(JSON.stringify(response));

View Console Output in your Browser

Now that everything has been correctly configured (fingers crossed) you can run your sample application and connect to Pieces OS for the first time.

Inside your terminal at the root directory of your project, use NPM to run one of the scripts that we added to the package.json file called "start":

npm run start

And you should have the same content in the main browser window as before once this completes. But if you open up your inspector using cmd + option + i or ctrl + shift + c you will see this inside of your console:

Console Output

This includes both the full OS response object with all the data that you will need to get going with other endpoints, and your application is now connected and ready to go for the rest of your exploration and discovery!

Follow along with these next steps to learn about assets and formats, two things that are very important for managing any form of data with Pieces OS.

Getting Started with Pieces.AssetApi + Pieces.AssetsApi

First lets define what these two things are:

  • Asset: a very important models who's primary purpose is to manage the seeded data that comes in to the application, and is stored inside of Pieces OS. Each asset is a identifiable piece of saved data, or pre-seeded data.

  • Assets: equally important, but instead of containing a single asset with parameters storing data on it, Assets serves as the list of type: Asset objects that are stored there. Also you will find the operations for adding, deleting, searching, and other functions that are related to referencing a number of different snippets to make a comparison. For instance:

If I want to create a snippet (lets call it var), I need to send this to the master Assets list, you would first create var itself with the proper formats and data added to the var object, then send the newly created SeededAsset over to the Assets list (assets/create). Then you will receive the asset back as the response from the server. Cool, right?

Traditionally, Assets is a linear list of flat Asset objects stored in an array or list.

You can use identifiers to get a specific asset from the asset list called a UUID. But you'll learn more about that later on.


Initially when creating your application, you will have no snippets saved to your project, will not be signed in, and you will have not completed onboarding. These are properties that you may see during this creation.

Check out localhost:1000/assets while Pieces OS is running to see the empty object that is there.

Creating your First Asset

While creating an asset, there are some required parameters that you will need to be sure to include the proper format data. For each Asset object, each required parameter must be included, and the Asset must be seeded before it is sent to be created.


This seed data will become the asset. You can use this structure to provide data to Pieces OS, and will include fewer parameters than what you will get back in your response. Lets get started with the seeded asset formatting before we pass this over to /Assets.

At the top level of this object you will see:

  • schema
  • metadata
  • application (required)
  • format (required)
  • discovered
  • available

Schema, metadata, discovered and available are all parameters that have extensive use cases, but for now we are going to focus on application & format - the two required formats for this object.

With each call you need to include your application object that you created earlier - and we can do this inside of the .then() following the return from connect() which is defined here:

connect().then(__ => {
    // parsing this value and restringifying it properly gives it its type.
    full_context = __;
    let _t = JSON.parse(JSON.stringify(full_context));
    applicationData = _t.application;
    console.log('Application Data: ', applicationData);



Now before continuing forward, we will need to prepare the create() function to connect to the proper creation endpoint. Create differs from connect, since previously our json object did not require any preprocessing. In this case we will need to include the application data that was returned back from our initial call to connect().

The createAsset() function needs to accomplish:

  1. Create our raw data var for seeding the asset.
  2. Creating a new asset using our simple Pieces.SeededAsset configuration
  3. Send request via Pieces.AssetsApi().assetsCreateNewAsset()
  4. Return the created asset back after it is validated and created

Here is what the createAsset() function looks like in its entirety:

// importing the package into this file.
import * as pieces from '@pieces.app/pieces-os-client'

// @var code data as a string.
var data = "<h1>Hello world</h1>";

// @var title for your snippet creation.
var name = "My First Snippet";

// the create asset function where we create our seeded asset.
// @var applicationData | look back at connect() to see where this came from
function createAsset() {
  let _seededAsset: Pieces.SeededAsset = {
    application: applicationData,
    format: {
      fragment: {
        string: {raw: data},
    metadata: {
      name: name

  // create your seed
  let _seed: Pieces.Seed = {
    asset: _seededAsset,
    type: SeedTypeEnum.Asset

  // make your api call.
  new Pieces.AssetsApi().assetsCreateNewAsset({seed: _seed}).then(newAsset => {
    console.log(`New Asset Created --> ${newAsset}`);

The response back will look similar to the following: https://jwaf.pieces.cloud/?p=24e242a85e

Get Assets Snapshot Pieces.AssetsApi().assetsSnapshot()

When reading along, if you would like to view your data incrementally through the full browser window, you can navigate to http://localhost:1000/assets to view a full list of snippets that have been saved in your browser. Otherwise, you can access the snapshot with these steps:

new Pieces.AssetsApi().assetsSnapshot({}).then(_assetList => {
    for (let i = 0; i < _assetList.iterable.length; i++) {
        // will log each asset.

Creating a Copilot

The following examples show how to use the Pieces Copilot and some of the endpoints available. Read more about the copilot on this article: Creating your own Open Source Copilot.

Use Pieces.QGPTStreamInput

You can query the Pieces Copilot out of the box (after installation) with no application context set. The stream input is structured as a question object, containing a query and a parameter for relevance called relevant:

const input: Pieces.QGPTStreamInput = {
	question: {
		relevant: {iterable: []}

Then you can use a number of stream listeners to inside of something like a websocket like seen here to deal with any JSON configuration.

Receiving Answers with Pieces.QGPTQuestionAnswer

When you get a response back from the copilot after asking a question, you may need to type it accordingly to access the appropriate properties. Here would be an example if getting back the first answer message following a question that was sent over:

const answer: Pieces.QGPTQuestionAnswer | undefined = result.question?.answers.iterable[0];

Read more on the specifics of the Pieces Copilot logic and endpoints on this copilot specific repo.

Each asset will have and ID on it that can be used to match a singular asset here. Very useful when trying to get a specific asset from your full assetsSnapshot.

Using Pieces.AssetApi().assetUpdate() to Rename Snippet

Individual assets can be manipulated with a number of different properties and metadata. You can add titles, annotations, tags, links, anchors, and much more all through this single endpoint. To use it properly first use the assetSnapshot to get your asset using its ID property, and store your asset in a variable in your .then(). You can then adjust any of the properties on this asset you have stored on _asset, then pass into requestParameters on the assetUpdate endpoint.

Check out this code block with an example of how to rename an asset:

function renameAsset(_name: string, _id: String){
    new Pieces.AssetsApi().assetsSnapshot({}).then(_assetList => {
        for (let i = 0; i < _assetList.iterable.length; i++) {
            if (_assetList.iterable[i].id == _id) {

                let _asset = _assetList.iterable[i];

                _asset.name = _name;

                new Pieces.AssetApi().assetUpdate({asset: _asset}).then(_updated => {
                    console.log("updated:", _updated);

Then inside your UI you can connect the renameAsset() function to your button and text inputs. Note that in order to perform this operation you need both the id of the asset, and also the new name that you would like to update you asset to.

The code block below includes the state management for the text input, along with the button that initiates the asset update as well.

This is what that looks like:

function RenameAssetInput() {
    const [name, setNameValue] = useState('');
    const [id, setIdValue] = useState('');

    const handleNameChange = (event) => {

    const handleIdChange = (event) => {

    return (
            <input value={name} style={{ width: '450px', verticalAlign: 'top' }} onChange={handleNameChange} />
            <input value={id} style={{ width: '450px', verticalAlign: 'top' }} onChange={handleIdChange} />
            <button style={{ marginTop: '10px', maxWidth: '200px' }} onClick={() => renameAsset(name, id)}>Rename Snippet</button>


Refresh Your Snapshot

In order to get updates to your assetSnapshot as a whole, you may need to update you local list in order to reflect changes that come from Pieces OS and give information on the assets stored there. In order to perform a refresh you can use this code block here:

const [array, setArray] = useState([]);

const refresh = (_newAsset: LocalAsset) => {
    setArray(prevArray => [...prevArray, _newAsset])

function refreshSnippetList() {
    new Pieces.AssetsApi().assetsSnapshot({}).then((assets) => {
        // loop through your assets.
        for (let i = 0; i < assets.iterable.length; i++) {
            let _local: LocalAsset = {
                id: assets.iterable[i].id,
                name: assets.iterable[i].name,
                classification: assets.iterable[i].original.reference.classification.specific



I added this to the top level for reactivity inside the main App() call. You can choose to place this in a different location if you are not in need of any reactive data.

Delete Using Pieces.AssetsApi().assetsDeleteAsset()

Assets can be deleted from your Assets list entirely by passing them into the assetsDeleteAsset endpoint. Just like the above example to rename a specific asset, you will need the ID of the asset that you are trying to remove. In order to get that you will need to use assetSnapshot in tandem with your delete endpoint:

 new Pieces.AssetsApi().assetsSnapshot({}).then(_assetList => {
    for (let i = 0; i < _assetList.iterable.length; i++) {
        if (_assetList.iterable[i].id == _id) {
            new Pieces.AssetsApi().assetsDeleteAsset({asset: _assetList.iterable[i].id }).then(_ => console.log("delete confirmed!"))

After a successful delete, you may have to reload your browser window in order to see the updated snippet list.

We use JSON Viewer internally when developing and recommend using some form of web based extension that assists with reading JSON DATA

Use Pieces.SearchApi()

The search API can be used to filter or search through snippets that have been saved, then perform specific actions on them based on a set of rules. Here is a brief example of searching where query: "page" is your search term:

new Pieces.SearchApi().fullTextSearch({ query: "page" }).then( searchedAssets => {

    // get the "ID" or identifier of the first match on the string you passed in as the query:
    let firstSearchMatchAssetIdentifier = searchedAssets.iterable[0].identifier;

    let matchName: String;

    // take that identifier to get your assets name using the Pieces.AssetApi()
    new Pieces.AssetApi().assetSnapshot({asset: firstSearchMatchAssetIdentifier}).then((asset) => {
      // assign that name to the matchName variable:
      matchName = asset.name;
    // then you can do whatever you would like with that match:   
  return matchName;


This is a very simple guide on how to get up and running using the @pieces.app/pieces-os-client npm package and create a web environment that you can build on top of along with examples on how to use some of the specific endpoints as well. Fork this repo to get started and learn about the depth of possibilities you have with Pieces OS.

To continue learning about the endpoints, other supported languages, and more documentation, check out the OpenSource repo.

Package Sidebar


npm i @pieces.app/pieces-os-client

Weekly Downloads






Unpacked Size

3.17 MB

Total Files


Last publish


  • robert-at-pieces
  • caleb-at-pieces
  • nathan-pieces
  • mack-at-pieces
  • mdwidman