Working with Content Archives

When you upload files to NFT.Storage using the client library, your data is converted into a graph of data structures, which are then packed into a format called a Content Archive (CAR) before being sent to the NFT.Storage service.

For many use cases, you never need to know about this process, as the conversion happens behind the scenes when using the client library. However, if you're using the HTTP API, or if you want more control over the structure of the IPFS data graph, you may want to work with Content Archives directly.

This how-to guide will explain the basics of Content Archives and how they're used by the NFT.Storage API.

We'll also see several methods of creating and manipulating Content Archives using command line tools and an overview of the libraries you can use in your application's code.

What is a Content Archive?

The Content Archive format is a way of packaging up [content addressed data][concepts-content-addressing] into archive files that can be easily stored and transferred. You can think of them like TAR files that are designed for storing collections of content addressed data.

The type of data stored in CARs is defined by IPLD, or InterPlanetary Linked Data. IPLD is a specification and set of implementations for structured data types that can link to each other using a hash-based Content Identifier (CID). Data linked in this way forms a Directed Acyclic Graph, or DAG, and you'll likely see a few references to DAGs in the documentation for IPLD and IPFS.

IPFS files are one example of IPLD data, but IPLD can also be used to access data from Ethereum, Git, and other hash-addressed systems. You can also use IPLD as a general purpose format for your structured data, sort of like a Web3-flavored JSON. See Advanced IPLD formats below for more information.

CARs and NFT.Storage

When the NFT.Storage client packs up regular files into a CAR to store on IPFS, the CAR contains data encoded in the same format used by IPFS when importing files using the command line or other IPFS APIs.

This format uses an IPLD "codec" called dag-pb, which uses Protocol Buffers to encode an object graph. Inside the graph are UnixFS objects that describe the files and their contents.

Although the HTTP API also allows you to upload regular files, the client prefers to send CARs for a few reasons.

First, formatting everything on the client allows us to calculate the root Content Identifier for the data you're uploading before we send any data to the remote service. This means that you can compare the CID returned by the NFT.Storage service to the one you calculated locally, and you don't have to trust the service to do the right thing.

Another reason to use CARs is to support large files, which would otherwise hit size limits on the NFT.Storage backend platform. The data in a CAR is already chunked into small blocks, which makes CARs easy to split into small pieces that can be uploaded in batches.

Command line tools

There are a few ways to create and interact with CAR files from the command line.


The ipfs-car JavaScript package includes a command-line tool for easily creating, unpacking, and verifying CAR files.

To install it, you'll need Node.js - we recommend the latest stable version.

You can install the command globally:

npm install -g ipfs-car

Or run the command with npx without installing it to your PATH:

npx ipfs-car --help

The --pack flag will create a new CAR file from a collection of input files:

ipfs-car --pack path/to/files --output path/to/write/

Or extract files from a CAR with --unpack:

ipfs-car --unpack path/to/ --output /path/to/unpack/files/to

You can also list the contents of a CAR with --list:

ipfs-car --list path/to/

For more usage information, run ipfs-car --help.


go-ipfs is the reference implementation of the IPFS protocol. Among many other features, go-ipfs supports exporting any IPFS object graph into a CAR file and importing data from CAR files into your local IPFS repository.

The ipfs dag export command will fetch an IPFS object graph by its Content ID (CID), writing a stream of CAR data to standard output.

To create a CAR file using go-ipfs, you can redirect the output of ipfs dag export to a file:

ipfs dag export $cid > path/to/

Note that you should replace the value of cid inside the quotes with the CID you want to export.

If you don't have the CID in your local IPFS repository, the dag export command will try to fetch it over the IPFS network.

To add the contents of a CAR file to your local IPFS repository, you can use ipfs dag import:

ipfs dag import path/to/

Libraries for application developers


There are two JavaScript packages available for manipulating CARs inside your application.


The ipfs-car package includes library functions for packing and unpacking files into CARs, using the IPFS UnixFs data model. The library includes the same functionality as the ipfs-car command line utility described above.

See the ipfs-car README for API documentation and usage examples.


The @ipld/car package contains the main JavaScript implementation of the CAR specification and is used by ipfs-car under the hood. If you want to store non-file data using advanced IPLD formats, you should use @ipld/car directly.

@ipld/car also provides the CarReader interface used by the NFT.Storage client's storeCar method.

Here's a simple example of loading a CAR file from a Node.js stream and storing it with NFT.Storage:

import { createReadStream } from 'fs'
import { CarReader } from '@ipld/car'
async function storeCarFile(filename) {
const inStream = createReadStream(filename)
const car = await CarReader.fromIterable(inStream)
const client = makeStorageClient()
const cid = await client.putCar(car)
console.log('Stored CAR file! CID:', cid)

CarReader.fromIterable accepts any iterable of Uint8Array data, including Node.js streams. If you have all your CAR data in a single Uint8Array already, you can use CarReader.fromBytes instead.

The CarReader type shown above will read the entire contents of the CAR into memory, which may cause issues with large files. On Node.js, you can use CarIndexedReader, which reads CAR data from disk directly and uses less memory than CarReader.


The go-car module provides the main Golang implementation of the CAR specification. We recommend using the v2 module version, which supports the latest version of the CAR spec.

See the API reference documentation for more information.

Splitting CARs for upload to NFT.Storage

The NFT.Storage HTTP API accepts CAR uploads up to 100 MB in size, but the JavaScript client uses the HTTP API to upload files of any size. The client manages to do this by splitting CARs into chunks of less than 100 MB each and uploading each chunk separately.

The main tool available for splitting and joining CARs is called carbites, which has implementations in JavaScript and Go. The JavaScript implementation includes a command-line version that allows you to split and join CARs from your terminal or favorite scripting language.

This section will demonstrate a few ways to split CARs in a way that's acceptable to the NFT.Storage service, using the command line tool, as well as programmatically using the carbites libraries in JavaScript and Go.

Using the carbites-cli tool

The JavaScript carbites library includes a package called carbites-cli that can split and join CARs from the command line. You'll need a recent version of Node.js installed, preferably the latest stable version.

You can install the tool globally with npm:

npm install -g carbites-cli
added 71 packages, and audited 72 packages in 846ms
20 packages are looking for funding
  run `npm fund` for details
found 0 vulnerabilities

This will add a carbites command to your shell's environment:

carbites --help
  CLI tool for splitting a single CAR into multiple CARs from the comfort of your terminal.
    $ carbites <command>

You can run the carbites command without installing it globally using the npx command, which is included with Node.js:

npx carbites-cli --help

The first time around, it will ask to make sure you want to install the package:

Need to install the following packages:
Ok to proceed? (y)

After that, you can use npx carbites-cli instead of carbites for any of the commands below!

Splitting CARs

The carbites split command takes a CAR file as input and splits it into multiple smaller CARs.

The --size flag sets the maximum size of the output CAR files. For uploading to NFT.Storage, --size must be less than 100MB.

The other important flag is --strategy, which determines how the CAR files are split. For NFT.Storage uploads, we need to use the treewalk strategy, so that all of our CARs share the same root CID. This will allow the NFT.Storage service to piece them all together again once they've all been uploaded.

Here's an example, using an input car file called that weighs in at 455MB:

carbites split --size 100MB --strategy treewalk

This will create five new files in the same directory as the input file, named through If you list their sizes, you can see that all the chunked cars are less than or equal to 100 MB:

ls -lh my-video*
-rw-r--r--  1 user  staff   100M Sep 15 13:56
-rw-r--r--  1 user  staff   100M Sep 15 13:56
-rw-r--r--  1 user  staff   100M Sep 15 13:56
-rw-r--r--  1 user  staff   100M Sep 15 13:56
-rw-r--r--  1 user  staff    56M Sep 15 13:56
-rw-r--r--  1 user  staff   455M Sep 15 13:52

Joining CARs

To combine CARs that have been previously split, you can use the carbites join command:

carbites join my-video-*.car --output

Using JavaScript code

The carbites library provides an interface for splitting CARs that can be invoked from your application code.


You probably don't need this!

If you're using JavaScript, you can use the NFT.Storage client to upload your data and let the client take care of CAR splitting for you. If you're sure you want to split CARs from JavaScript yourself, read on!

To split CARs from your JavaScript code, install the carbites package:

npm install carbites

And import the TreewalkCarSplitter class into your code:

import { TreewalkCarSplitter } from 'carbites/treewalk'

You can create a TreewalkCarSplitter by passing in a CarReader and a targetSize in bytes for the output cars. See the section on @ipld/car for more information on CarReader. For now, we'll assume that the loadLargeCar function returns a CarReader, and we'll use the TreewalkCarSplitter to create split CARs:

import { TreewalkCarSplitter } from 'carbites/treewalk'
async function splitCars() {
const largeCar = await loadLargeCar()
const targetSize = 100000000
const splitter = new TreewalkCarSplitter(largeCar, targetSize)
for await (const smallCar of {
// Each small car is an AsyncIterable<Uint8Array> of CAR data
for await (const chunk of smallCar) {
// Do something with the car data...
// For example, you could upload it to the HTTP API
// You can also get the root CID of each small CAR with the getRoots method:
const roots = await smallCar.getRoots()
console.log('root cids', roots)
// Since we're using TreewalkCarSpliter, all the smaller CARs should have the
// same root CID as the large input CAR.

Using Go code

The go-carbites module can be used to split large CARs from your Go applications.

Install the module with go get:

go get

The carbites.Split function returns a carbites.Splitter that will make sure that the output CARs all have the same root CID, which is important when uploading to NFT.Storage.

package main
import (
func main() {
bigCar, _ := os.Open("")
targetSize := 1024 * 1024 // 1MiB chunks
strategy := carbites.Treewalk
spltr, _ := carbites.Split(bigCar, targetSize, strategy)
var i int
for {
car, err := spltr.Next()
if err != nil {
if err == io.EOF {
b, _ := ioutil.ReadAll(car)
ioutil.WriteFile(fmt.Sprintf("", i), b, 0644)

You can also use NewTreewalkSplitterFromPath, which takes a local file path instead of an io.Reader.

Advanced IPLD formats

IPLD can also be used as a general purpose data format like JSON. In fact, you can use JSON directly as IPLD just by using a special convention for linking to other IPLD objects. This convention is defined in the dag-json "codec".

Here's an example of a dag-json object:

"name": "Have you seen this dog?",
"description": "I have now...",
"image": { "/": "bafybeihkqv2ukwgpgzkwsuz7whmvneztvxglkljbs3zosewgku2cfluvba" }

The image field uses the special "link type" to reference another IPLD object. The link is just a regular JSON object with a single key named /, whose value is a Content Identifier.

Although dag-json is familiar and easy to use, we recommend using the similar dag-cbor codec instead. dag-cbor uses the Concise Binary Object Representation to more efficiently encode data, especially binary data which must be Base64-encoded when using dag-json.

You may already be using dag-cbor without knowing it! When using the client library's store method or the HTTP API's /store endpoint, NTF.Storage creates a dag-cbor "bundle" that includes the NFT metadata with native IPLD links to all of the files referenced in the NFT, including a JSON version of the metadata for compatibility.