Bucket Node.js SDK

Node.js, JavaScript/TypeScript client for Bucket.co.

Bucket supports feature toggling, tracking feature usage, collecting feedback on features, and remotely configuring features.

Installation

Install using your favorite package manager:

{% tabs %} {% tab title="npm" %}

1npm i @bucketco/node-sdk

{% endtab %}

{% tab title="yarn" %}

1yarn add @bucketco/node-sdk

{% endtab %}

{% tab title="bun" %}

1bun add @bucketco/node-sdk

{% endtab %}

{% tab title="pnpm" %}

1pnpm add @bucketco/node-sdk

{% endtab %}

{% tab title="deno" %}

1deno add npm:@bucketco/node-sdk

{% endtab %} {% endtabs %}

Other supported languages/frameworks are in the Supported languages documentation pages.

You can also use the HTTP API directly

Basic usage

To get started you need to obtain your secret key from the environment settings in Bucket.

[!CAUTION] Secret keys are meant for use in server side SDKs only. Secret keys offer the users the ability to obtain information that is often sensitive and thus should not be used in client-side applications.

Bucket will load settings through the various environment variables automatically (see Configuring below).

1. Find the Bucket secret key for your development environment under environment settings in Bucket.
2. Set BUCKET_SECRET_KEY in your .env file
3. Create a bucket.ts file containing the following:

1import { BucketClient } from "@bucketco/node-sdk";
2
3// Create a new instance of the client with the secret key. Additional options
4// are available, such as supplying a logger and other custom properties.
5//
6// We recommend that only one global instance of `client` should be created
7// to avoid multiple round-trips to our servers.
8export const bucketClient = new BucketClient();
9
10// Initialize the client and begin fetching feature targeting definitions.
11// You must call this method prior to any calls to `getFeatures()`,
12// otherwise an empty object will be returned.
13bucketClient.initialize().then({
14  console.log("Bucket initialized!")
15})

Once the client is initialized, you can obtain features along with the isEnabled status to indicate whether the feature is targeted for this user/company:

[!IMPORTANT] If user.id or company.id is not given, the whole user or company object is ignored.

1// configure the client
2const boundClient = bucketClient.bindClient({
3  user: {
4    id: "john_doe",
5    name: "John Doe",
6    email: "john@acme.com",
7    avatar: "https://example.com/users/jdoe",
8  },
9  company: {
10    id: "acme_inc",
11    name: "Acme, Inc.",
12    avatar: "https://example.com/companies/acme",
13  },
14});
15
16// get the huddle feature using company, user and custom context to
17// evaluate the targeting.
18const { isEnabled, track, config } = boundClient.getFeature("huddle");
19
20if (isEnabled) {
21  // this is your feature gated code ...
22  // send an event when the feature is used:
23  track();
24
25  if (config?.key === "zoom") {
26    // this code will run if a given remote configuration
27    // is set up.
28  }
29
30  // CAUTION: if you plan to use the event for automated feedback surveys
31  // call `flush` immediately after `track`. It can optionally be awaited
32  // to guarantee the sent happened.
33  boundClient.flush();
34}

You can also use the getFeatures() method which returns a map of all features:

1// get the current features (uses company, user and custom context to
2// evaluate the features).
3const features = boundClient.getFeatures();
4const bothEnabled =
5  features.huddle?.isEnabled && features.voiceHuddle?.isEnabled;

High performance feature targeting

The SDK contacts the Bucket servers when you call initialize() and downloads the features with their targeting rules. These rules are then matched against the user/company information you provide to getFeatures() (or through bindClient(..).getFeatures()). That means the getFeatures() call does not need to contact the Bucket servers once initialize() has completed. BucketClient will continue to periodically download the targeting rules from the Bucket servers in the background.

Batch Operations

The SDK automatically batches operations like user/company updates and feature tracking events to minimize API calls. The batch buffer is configurable through the client options:

1const client = new BucketClient({
2  batchOptions: {
3    maxSize: 100, // Maximum number of events to batch
4    intervalMs: 1000, // Flush interval in milliseconds
5  },
6});

You can manually flush the batch buffer at any time:

1await client.flush();

[!TIP] It's recommended to call flush() before your application shuts down to ensure all events are sent.

Rate Limiting

The SDK includes automatic rate limiting for feature events to prevent overwhelming the API. Rate limiting is applied per unique combination of feature key and context. The rate limiter window size is configurable:

1const client = new BucketClient({
2  rateLimiterOptions: {
3    windowSizeMs: 60000, // Rate limiting window size in milliseconds
4  },
5});

Feature definitions

Feature definitions include the rules needed to determine which features should be enabled and which config values should be applied to any given user/company. Feature definitions are automatically fetched when calling initialize(). They are then cached and refreshed in the background. It's also possible to get the currently in use feature definitions:

1import fs from "fs";
2
3const client = new BucketClient();
4
5const featureDefs = await client.getFeatureDefinitions();
6// [{
7//   key: "huddle",
8//   description: "Live voice conversations with colleagues."
9//   flag: { ... }
10//   config: { ... }
11// }]

Edge-runtimes like Cloudflare Workers

To use the Bucket NodeSDK with Cloudflare workers, set the node_compat flag in your wrangler file.

Instead of using BucketClient, use EdgeClient and make sure you call ctx.waitUntil(bucket.flush()); before returning from your worker function.

1import { EdgeClient } from "@bucketco/node-sdk";
2
3// set the BUCKET_SECRET_KEY environment variable or pass the secret key in the constructor
4const bucket = new EdgeClient();
5
6export default {
7  async fetch(request, _env, ctx): Promise<Response> {
8    // initialize the client and wait for it to complete
9    // if the client was initialized on a previous invocation, this is a no-op.
10    await bucket.initialize();
11    const features = bucket.getFeatures({
12      user: { id: "userId" },
13      company: { id: "companyId" },
14    });
15
16    // ensure all events are flushed and any requests to refresh the feature cache
17    // have completed after the response is sent
18    ctx.waitUntil(bucket.flush());
19
20    return new Response(
21      `Features for user ${userId} and company ${companyId}: ${JSON.stringify(features, null, 2)}`,
22    );
23  },
24};

See examples/cloudflare-worker for a deployable example.

Bucket maintains a cached set of feature definitions in the memory of your worker which it uses to decide which features to turn on for which users/companies.

The SDK caches feature definitions in memory for fast performance. The first request to a new worker instance fetches definitions from Bucket's servers, while subsequent requests use the cache. When the cache expires, it's updated in the background. ctx.waitUntil(bucket.flush()) ensures completion of the background work, so response times are not affected. This background work may increase wall-clock time for your worker, but it will not measurably increase billable CPU time on platforms like Cloudflare.

Error Handling

The SDK is designed to fail gracefully and never throw exceptions to the caller. Instead, it logs errors and provides fallback behavior:

1. Feature Evaluation Failures:

   1const { isEnabled } = client.getFeature("my-feature");
   2// If feature evaluation fails, isEnabled will be false

2. Network Errors:

   1// Network errors during tracking are logged but don't affect your application
   2const { track } = client.getFeature("my-feature");
   3if (isEnabled) {
   4  try {
   5    await track();
   6  } catch (error) {
   7    // The SDK already logged this error
   8    // Your application can continue normally
   9  }
   10}

3. Missing Context:

   1// The SDK tracks missing context fields but continues operation
   2const features = client.getFeatures({
   3  user: { id: "user123" },
   4  // Missing company context will be logged but won't cause errors
   5});

4. Offline Mode:

   1// In offline mode, the SDK uses feature overrides
   2const client = new BucketClient({
   3  offline: true,
   4  featureOverrides: () => ({
   5    "my-feature": true,
   6  }),
   7});

The SDK logs all errors with appropriate severity levels. You can customize logging by providing your own logger:

1const client = new BucketClient({
2  logger: {
3    debug: (msg) => console.debug(msg),
4    info: (msg) => console.info(msg),
5    warn: (msg) => console.warn(msg),
6    error: (msg, error) => {
7      console.error(msg, error);
8      // Send to your error tracking service
9      errorTracker.capture(error);
10    },
11  },
12});

Remote config

Remote config is a dynamic and flexible approach to configuring feature behavior outside of your app – without needing to re-deploy it.

Similar to isEnabled, each feature has a config property. This configuration is managed from within Bucket. It is managed similar to the way access to features is managed, but instead of the binary isEnabled you can have multiple configuration values which are given to different user/companies.

1const features = bucketClient.getFeatures();
2// {
3//   huddle: {
4//     isEnabled: true,
5//     targetingVersion: 42,
6//     config: {
7//       key: "gpt-3.5",
8//       payload: { maxTokens: 10000, model: "gpt-3.5-beta1" }
9//     }
10//   }
11// }

key is mandatory for a config, but if a feature has no config or no config value was matched against the context, the key will be undefined. Make sure to check against this case when trying to use the configuration in your application. payload is an optional JSON value for arbitrary configuration needs.

Just as isEnabled, accessing config on the object returned by getFeatures does not automatically generate a check event, contrary to the config property on the object returned by getFeature.

Configuring

The Bucket Node.js SDK can be configured through environment variables, a configuration file on disk or by passing options to the BucketClient constructor. By default, the SDK searches for bucketConfig.json in the current working directory.

Option	Type	Description	Env Var
secretKey	string	The secret key used for authentication with Bucket's servers.	BUCKET_SECRET_KEY
logLevel	string	The log level for the SDK (e.g., "DEBUG", "INFO", "WARN", "ERROR"). Default: INFO	BUCKET_LOG_LEVEL
offline	boolean	Operate in offline mode. Default: false, except in tests it will default to true based off of the TEST env. var.	BUCKET_OFFLINE
apiBaseUrl	string	The base API URL for the Bucket servers.	BUCKET_API_BASE_URL
featureOverrides	Record<string, boolean>	An object specifying feature overrides for testing or local development. See examples/express/app.test.ts for how to use featureOverrides in tests.	BUCKET_FEATURES_ENABLED, BUCKET_FEATURES_DISABLED
configFile	string	Load this config file from disk. Default: bucketConfig.json	BUCKET_CONFIG_FILE

[!NOTE] > BUCKET_FEATURES_ENABLED and BUCKET_FEATURES_DISABLED are comma separated lists of features which will be enabled or disabled respectively.

bucketConfig.json example:

1{
2  "secretKey": "...",
3  "logLevel": "warn",
4  "offline": true,
5  "apiBaseUrl": "https://proxy.slick-demo.com",
6  "featureOverrides": {
7    "huddles": true,
8    "voiceChat": { "isEnabled": false },
9    "aiAssist": {
10      "isEnabled": true,
11      "config": {
12        "key": "gpt-4.0",
13        "payload": {
14          "maxTokens": 50000
15        }
16      }
17    }
18  }
19}

When using a bucketConfig.json for local development, make sure you add it to your .gitignore file. You can also set these options directly in the BucketClient constructor. The precedence for configuration options is as follows, listed in the order of importance:

1. Options passed along to the constructor directly,
2. Environment variable,
3. The config file.

Type safe feature flags

To get type checked feature flags, install the Bucket CLI:

npm i --save-dev @bucketco/cli

then generate the types:

npx bucket features types

This will generate a bucket.d.ts containing all your features. Any feature look ups will now be checked against the features that exist in Bucket.

Here's an example of a failed type check:

1import { BucketClient } from "@bucketco/node-sdk";
2
3export const bucketClient = new BucketClient();
4
5bucketClient.initialize().then(() => {
6  console.log("Bucket initialized!");
7
8  // TypeScript will catch this error: "invalid-feature" doesn't exist
9  bucketClient.getFeature("invalid-feature");
10
11  const {
12    isEnabled,
13    config: { payload },
14  } = bucketClient.getFeature("create-todos");
15});

This is an example of a failed config payload check:

1bucketClient.initialize().then(() => {
2  // TypeScript will catch this error as well: "minLength" is not part of the payload.
3  if (isEnabled && todo.length > config.payload.minLength) {
4    // ...
5  }
6});

Testing

When writing tests that cover code with feature flags, you can toggle features on/off programmatically to test the different behavior.

bucket.ts:

1import { BucketClient } from "@bucketco/node-sdk";
2
3export const bucket = new BucketClient();

app.test.ts:

1import { bucket } from "./bucket.ts";
2
3beforeAll(async () => await bucket.initialize());
4afterEach(() => {
5  bucket.clearFeatureOverrides();
6});
7
8describe("API Tests", () => {
9  it("should return 200 for the root endpoint", async () => {
10    bucket.featureOverrides = {
11      "show-todo": true,
12    };
13
14    const response = await request(app).get("/");
15    expect(response.status).toBe(200);
16    expect(response.body).toEqual({ message: "Ready to manage some TODOs!" });
17  });
18});

See more on feature overrides in the section below.

Feature Overrides

Feature overrides allow you to override feature flags and their configurations locally. This is particularly useful for development and testing. You can specify overrides in three ways:

1. Through environment variables:

1BUCKET_FEATURES_ENABLED=feature1,feature2
2BUCKET_FEATURES_DISABLED=feature3,feature4

2. Through bucketConfig.json:

1{
2  "featureOverrides": {
3    "delete-todos": {
4      "isEnabled": true,
5      "config": {
6        "key": "dev-config",
7        "payload": {
8          "requireConfirmation": true,
9          "maxDeletionsPerDay": 5
10        }
11      }
12    }
13  }
14}

3. Programmatically through the client options:

You can use a simple Record<string, boolean> and pass it either in the constructor or by setting client.featureOverrides:

1// pass directly in the constructor
2const client = new BucketClient({ featureOverrides: { myFeature: true } });
3// or set on the client at a later time
4client.featureOverrides = { myFeature: false };
5
6// clear feature overrides. Same as setting to {}.
7client.clearFeatureOverrides();

To get dynamic overrides, use a function which takes a context and returns a boolean or an object with the shape of {isEnabled, config}:

1import { BucketClient, Context } from "@bucketco/node-sdk";
2
3const featureOverrides = (context: Context) => ({
4  "delete-todos": {
5    isEnabled: true,
6    config: {
7      key: "dev-config",
8      payload: {
9        requireConfirmation: true,
10        maxDeletionsPerDay: 5,
11      },
12    },
13  },
14});
15
16const client = new BucketClient({
17  featureOverrides,
18});

Remote Feature Evaluation

In addition to local feature evaluation, Bucket supports remote evaluation using stored context. This is useful when you want to evaluate features using user/company attributes that were previously sent to Bucket:

1// First, update user and company attributes
2await client.updateUser("user123", {
3  attributes: {
4    role: "admin",
5    subscription: "premium",
6  },
7});
8
9await client.updateCompany("company456", {
10  attributes: {
11    tier: "enterprise",
12    employees: 1000,
13  },
14});
15
16// Later, evaluate features remotely using stored context
17const features = await client.getFeaturesRemote("company456", "user123");
18// Or evaluate a single feature
19const feature = await client.getFeatureRemote(
20  "create-todos",
21  "company456",
22  "user123",
23);
24
25// You can also provide additional context
26const featuresWithContext = await client.getFeaturesRemote(
27  "company456",
28  "user123",
29  {
30    other: {
31      location: "US",
32      platform: "mobile",
33    },
34  },
35);

Remote evaluation is particularly useful when:

• You want to use the most up-to-date user/company attributes stored in Bucket
• You don't want to pass all context attributes with every evaluation
• You need to ensure consistent feature evaluation across different services

Using with Express

A popular way to integrate the Bucket Node.js SDK is through an express middleware.

1import bucket from "./bucket";
2import express from "express";
3import { BoundBucketClient } from "@bucketco/node-sdk";
4
5// Augment the Express types to include a `boundBucketClient` property on the
6// `res.locals` object.
7// This will allow us to access the BucketClient instance in our route handlers
8// without having to pass it around manually
9declare global {
10  namespace Express {
11    interface Locals {
12      boundBucketClient: BoundBucketClient;
13    }
14  }
15}
16
17// Add express middleware
18app.use((req, res, next) => {
19  // Extract the user and company IDs from the request
20  // You'll want to use a proper authentication and identification
21  // mechanism in a real-world application
22  const user = {
23    id: req.user?.id,
24    name: req.user?.name
25    email: req.user?.email
26  }
27
28  const company = {
29    id: req.user?.companyId
30    name: req.user?.companyName
31  }
32
33  // Create a new BoundBucketClient instance by calling the `bindClient`
34  // method on a `BucketClient` instance
35  // This will create a new instance that is bound to the user/company given.
36  const boundBucketClient = bucket.bindClient({ user, company });
37
38  // Store the BoundBucketClient instance in the `res.locals` object so we
39  // can access it in our route handlers
40  res.locals.boundBucketClient = boundBucketClient;
41  next();
42});
43
44// Now use res.locals.boundBucketClient in your handlers
45app.get("/todos", async (_req, res) => {
46  const { track, isEnabled } = res.locals.bucketUser.getFeature("show-todos");
47
48  if (!isEnabled) {
49    res.status(403).send({"error": "feature inaccessible"})
50    return
51  }
52
53  ...
54}

See examples/express/app.ts for a full example.

Remote flag evaluation with stored context

If you don't want to provide context each time when evaluating feature flags but rather you would like to utilize the attributes you sent to Bucket previously (by calling updateCompany and updateUser) you can do so by calling getFeaturesRemote (or getFeatureRemote for a specific feature) with providing just userId and companyId. These methods will call Bucket's servers and feature flags will be evaluated remotely using the stored attributes.

1// Update user and company attributes
2client.updateUser("john_doe", {
3  attributes: {
4    name: "John O.",
5    role: "admin",
6  },
7});
8
9client.updateCompany("acme_inc", {
10  attributes: {
11    name: "Acme, Inc",
12    tier: "premium"
13  },
14});
15...
16
17// This will evaluate feature flags with respecting the attributes sent previously
18const features = await client.getFeaturesRemote("acme_inc", "john_doe");

[!IMPORTANT] User and company attribute updates are processed asynchronously, so there might be a small delay between when attributes are updated and when they are available for evaluation.

Opting out of tracking

There are use cases in which you not want to be sending user, company and track events to Bucket.co. These are usually cases where you could be impersonating another user in the system and do not want to interfere with the data being collected by Bucket.

To disable tracking, bind the client using bindClient() as follows:

1// binds the client to a given user and company and set `enableTracking` to `false`.
2const boundClient = client.bindClient({ user, company, enableTracking: false });
3
4boundClient.track("some event"); // this will not actually send the event to Bucket.
5
6// the following code will not update the `user` nor `company` in Bucket and will
7// not send `track` events either.
8const { isEnabled, track } = boundClient.getFeature("user-menu");
9if (isEnabled) {
10  track();
11}

Another way way to disable tracking without employing a bound client is to call getFeature() or getFeatures() by supplying enableTracking: false in the arguments passed to these functions.

[!IMPORTANT] Note, however, that calling track(), updateCompany() or updateUser() in the BucketClient will still send tracking data. As such, it is always recommended to use bindClient() when using this SDK.

Flushing

BucketClient employs a batching technique to minimize the number of calls that are sent to Bucket's servers.

By default, the SDK automatically subscribes to process exit signals and attempts to flush any pending events. This behavior is controlled by the flushOnExit option in the client configuration:

1const client = new BucketClient({
2  batchOptions: {
3    flushOnExit: false, // disable automatic flushing on exit
4  },
5});

Tracking custom events and setting custom attributes

Tracking allows events and updating user/company attributes in Bucket. For example, if a customer changes their plan, you'll want Bucket to know about it, in order to continue to provide up-do-date targeting information in the Bucket interface.

The following example shows how to register a new user, associate it with a company and finally update the plan they are on.

1// registers the user with Bucket using the provided unique ID, and
2// providing a set of custom attributes (can be anything)
3client.updateUser("user_id", {
4  attributes: { longTimeUser: true, payingCustomer: false },
5});
6client.updateCompany("company_id", { userId: "user_id" });
7
8// the user started a voice huddle
9client.track("user_id", "huddle", { attributes: { voice: true } });

It's also possible to achieve the same through a bound client in the following manner:

1const boundClient = client.bindClient({
2  user: { id: "user_id", longTimeUser: true, payingCustomer: false },
3  company: { id: "company_id" },
4});
5
6boundClient.track("huddle", { attributes: { voice: true } });

Some attributes are used by Bucket to improve the UI, and are recommended to provide for easier navigation:

• name -- display name for user/company,
• email -- the email of the user,
• avatar -- the URL for user/company avatar image.

Attributes cannot be nested (multiple levels) and must be either strings, integers or booleans.

Managing Last seen

By default updateUser/updateCompany calls automatically update the given user/company Last seen property on Bucket servers.

You can control if Last seen should be updated when the events are sent by setting meta.active = false. This is often useful if you have a background job that goes through a set of companies just to update their attributes but not their activity.

Example:

1client.updateUser("john_doe", {
2  attributes: { name: "John O." },
3  meta: { active: true },
4});
5
6client.updateCompany("acme_inc", {
7  attributes: { name: "Acme, Inc" },
8  meta: { active: false },
9});

bindClient() updates attributes on the Bucket servers but does not automatically update Last seen.

Zero PII

The Bucket SDK doesn't collect any metadata and HTTP IP addresses are not being stored. For tracking individual users, we recommend using something like database ID as userId, as it's unique and doesn't include any PII (personal identifiable information). If, however, you're using e.g. email address as userId, but prefer not to send any PII to Bucket, you can hash the sensitive data before sending it to Bucket:

1import { sha256 } from 'crypto-hash';
2
3client.updateUser({ userId: await sha256("john_doe"), ... });

Typescript

Types are bundled together with the library and exposed automatically when importing through a package manager.

License

MIT License Copyright (c) 2025 Bucket ApS