diff --git a/accum/block.go b/accum/block.go
index e4861028..9c56c827 100644
--- a/accum/block.go
+++ b/accum/block.go
@@ -190,6 +190,17 @@ func (obj ObjectWithMetadata) RawSection() ([]byte, error) {
 	return buf, nil
 }
 
+func (obj ObjectWithMetadata) RawSectionSize() int {
+	sectionLen := len(obj.Cid.Bytes()) + len(obj.ObjectData)
+	lenBytes := leb128.FromUInt64(uint64(sectionLen))
+
+	// Size is:
+	// length of LEB128-encoded section length +
+	// length of CID bytes +
+	// length of object data
+	return len(lenBytes) + sectionLen
+}
+
 // {
 // 	raw, err := objm.RawSection()
 // 	if err != nil {
diff --git a/cmd-car-split.go b/cmd-car-split.go
new file mode 100644
index 00000000..3af2ce5a
--- /dev/null
+++ b/cmd-car-split.go
@@ -0,0 +1,326 @@
+package main
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"fmt"
+	"io"
+	"io/fs"
+	"os"
+
+	"github.com/filecoin-project/go-leb128"
+	"github.com/ipfs/go-cid"
+	"github.com/ipld/go-ipld-prime/codec/dagcbor"
+	"github.com/ipld/go-ipld-prime/datamodel"
+	"github.com/ipld/go-ipld-prime/fluent/qp"
+	cidlink "github.com/ipld/go-ipld-prime/linking/cid"
+	"github.com/ipld/go-ipld-prime/schema"
+	"github.com/multiformats/go-multicodec"
+	"github.com/rpcpool/yellowstone-faithful/accum"
+	"github.com/rpcpool/yellowstone-faithful/carreader"
+	"github.com/rpcpool/yellowstone-faithful/ipld/ipldbindcode"
+	"github.com/rpcpool/yellowstone-faithful/iplddecoders"
+	"github.com/urfave/cli/v2"
+	"k8s.io/klog/v2"
+)
+
+const (
+	nulRootCarHeader = "\x19" + // 25 bytes of CBOR (encoded as varint :cryingbear: )
+		// map with 2 keys
+		"\xA2" +
+		// text-key with length 5
+		"\x65" + "roots" +
+		// 1 element array
+		"\x81" +
+		// tag 42
+		"\xD8\x2A" +
+		// bytes with length 5
+		"\x45" +
+		// nul-identity-cid prefixed with \x00 as required in DAG-CBOR: https://github.com/ipld/specs/blob/master/block-layer/codecs/dag-cbor.md#links
+		"\x00\x01\x55\x00\x00" +
+		// text-key with length 7
+		"\x67" + "version" +
+		// 1, we call this v0 due to the nul-identity CID being an open question: https://github.com/ipld/go-car/issues/26#issuecomment-604299576
+		"\x01"
+)
+
+type subsetInfo struct {
+	fileName   string
+	firstSlot  int
+	lastSlot   int
+	blockLinks []datamodel.Link
+}
+
+func newCmd_SplitCar() *cli.Command {
+	return &cli.Command{
+		Name:        "split-car",
+		Description: "Splits an epoch car file into smaller chunks. Each chunk corresponds to a subset.",
+		ArgsUsage:   "<epoch-car-path>",
+		Flags: []cli.Flag{
+			&cli.Int64Flag{
+				Name:     "size",
+				Aliases:  []string{"s"},
+				Value:    31 * 1024 * 1024 * 1024, // 31 GiB
+				Usage:    "Target size in bytes to chunk CARs to.",
+				Required: false,
+			},
+			&cli.IntFlag{
+				Name:     "epoch",
+				Aliases:  []string{"e"},
+				Usage:    "Epoch number",
+				Required: true,
+			},
+		},
+		Action: func(c *cli.Context) error {
+			carPath := c.Args().First()
+			var file fs.File
+			var err error
+			if carPath == "-" {
+				file = os.Stdin
+			} else {
+				file, err = os.Open(carPath)
+				if err != nil {
+					return fmt.Errorf("failed to open CAR: %w", err)
+				}
+				defer file.Close()
+			}
+
+			rd, err := carreader.New(file)
+			if err != nil {
+				return fmt.Errorf("failed to open CAR: %w", err)
+			}
+			{
+				// print roots:
+				roots := rd.Header.Roots
+				klog.Infof("Roots: %d", len(roots))
+				for i, root := range roots {
+					if i == 0 && len(roots) == 1 {
+						klog.Infof("- %s (Epoch CID)", root.String())
+					} else {
+						klog.Infof("- %s", root.String())
+					}
+				}
+			}
+
+			epoch := c.Int("epoch")
+			maxFileSize := c.Int64("size")
+
+			var (
+				currentFileSize   int64
+				currentFileNum    int
+				currentFile       *os.File
+				bufferedWriter    *bufio.Writer
+				currentSubsetInfo subsetInfo
+				subsetLinks       []datamodel.Link
+			)
+
+			createNewFile := func() error {
+				if currentFile != nil {
+					sl, err := writeSubsetNode(currentSubsetInfo, bufferedWriter)
+					if err != nil {
+						return fmt.Errorf("failed to write subset node: %w", err)
+					}
+					subsetLinks = append(subsetLinks, sl)
+
+					err = closeFile(bufferedWriter, currentFile)
+					if err != nil {
+						return fmt.Errorf("failed to close file: %w", err)
+					}
+				}
+				currentFileNum++
+				filename := fmt.Sprintf("epoch-%d-%d.car", epoch, currentFileNum)
+				currentFile, err = os.Create(filename)
+				if err != nil {
+					return fmt.Errorf("failed to create file %s: %w", filename, err)
+				}
+
+				bufferedWriter = bufio.NewWriter(currentFile)
+
+				// Write the header
+				_, err = io.WriteString(bufferedWriter, nulRootCarHeader)
+				if err != nil {
+					return fmt.Errorf("failed to write header: %w", err)
+				}
+
+				// Set the currentFileSize to the size of the header
+				currentFileSize = int64(len(nulRootCarHeader))
+				currentSubsetInfo = subsetInfo{fileName: filename, firstSlot: -1, lastSlot: -1}
+				return nil
+			}
+
+			writeObject := func(data []byte) error {
+				_, err := bufferedWriter.Write(data)
+				if err != nil {
+					return fmt.Errorf("failed to write object to car file: %s, error: %w", currentFile.Name(), err)
+				}
+				currentFileSize += int64(len(data))
+				return nil
+			}
+
+			writeBlockDag := func(blockDag []accum.ObjectWithMetadata) error {
+				for _, owm := range blockDag {
+					rs, err := owm.RawSection()
+					if err != nil {
+						return fmt.Errorf("failed to get raw section: %w", err)
+					}
+
+					err = writeObject(rs)
+					if err != nil {
+						return fmt.Errorf("failed to write object: %w", err)
+					}
+				}
+
+				return nil
+			}
+
+			accum := accum.NewObjectAccumulator(
+				rd,
+				iplddecoders.KindBlock,
+				func(owm1 *accum.ObjectWithMetadata, owm2 []accum.ObjectWithMetadata) error {
+					if owm1 == nil {
+						return nil
+					}
+
+					owms := append(owm2, *owm1)
+					dagSize := 0
+
+					for _, owm := range owms {
+						dagSize += owm.RawSectionSize()
+					}
+
+					if currentFile == nil || currentFileSize+int64(dagSize) > maxFileSize {
+						err := createNewFile()
+						if err != nil {
+							return fmt.Errorf("failed to create a new file: %w", err)
+						}
+					}
+
+					// owm1 is necessarily a Block
+					block, err := iplddecoders.DecodeBlock(owm1.ObjectData)
+					if err != nil {
+						return fmt.Errorf("failed to decode block: %w", err)
+					}
+
+					if currentSubsetInfo.firstSlot == -1 || block.Slot < currentSubsetInfo.firstSlot {
+						currentSubsetInfo.firstSlot = block.Slot
+					}
+					if block.Slot > currentSubsetInfo.lastSlot {
+						currentSubsetInfo.lastSlot = block.Slot
+					}
+
+					currentSubsetInfo.blockLinks = append(currentSubsetInfo.blockLinks, cidlink.Link{Cid: owm1.Cid})
+
+					err = writeBlockDag(owms)
+					if err != nil {
+						return fmt.Errorf("failed to write block dag to file: %w", err)
+					}
+
+					return nil
+				},
+				iplddecoders.KindEpoch,
+				iplddecoders.KindSubset,
+			)
+
+			if err := accum.Run((context.Background())); err != nil {
+				return fmt.Errorf("failed to run accumulator while accumulating objects: %w", err)
+			}
+
+			sl, err := writeSubsetNode(currentSubsetInfo, bufferedWriter)
+			if err != nil {
+				return fmt.Errorf("failed to write subset node: %w", err)
+			}
+			subsetLinks = append(subsetLinks, sl)
+
+			epochNode, err := qp.BuildMap(ipldbindcode.Prototypes.Epoch, -1, func(ma datamodel.MapAssembler) {
+				qp.MapEntry(ma, "kind", qp.Int(int64(iplddecoders.KindEpoch)))
+				qp.MapEntry(ma, "epoch", qp.Int(int64(epoch)))
+				qp.MapEntry(ma, "subsets",
+					qp.List(-1, func(la datamodel.ListAssembler) {
+						for _, sl := range subsetLinks {
+							qp.ListEntry(la, qp.Link(sl))
+						}
+					}),
+				)
+			})
+			if err != nil {
+				return fmt.Errorf("failed to construct epochNode: %w", err)
+			}
+
+			_, err = writeNode(epochNode, bufferedWriter)
+			if err != nil {
+				return fmt.Errorf("failed to write epochNode: %w", err)
+			}
+
+			return closeFile(bufferedWriter, currentFile)
+		},
+	}
+}
+
+func writeSubsetNode(currentSubsetInfo subsetInfo, bufferedWriter *bufio.Writer) (datamodel.Link, error) {
+	subsetNode, err := qp.BuildMap(ipldbindcode.Prototypes.Subset, -1, func(ma datamodel.MapAssembler) {
+		qp.MapEntry(ma, "kind", qp.Int(int64(iplddecoders.KindSubset)))
+		qp.MapEntry(ma, "first", qp.Int(int64(currentSubsetInfo.firstSlot)))
+		qp.MapEntry(ma, "last", qp.Int(int64(currentSubsetInfo.lastSlot)))
+		qp.MapEntry(ma, "blocks",
+			qp.List(-1, func(la datamodel.ListAssembler) {
+				for _, bl := range currentSubsetInfo.blockLinks {
+					qp.ListEntry(la, qp.Link(bl))
+				}
+			}))
+	})
+	if err != nil {
+		return nil, fmt.Errorf("failed to write a subsetNode: %w", err)
+	}
+
+	cid, err := writeNode(subsetNode, bufferedWriter)
+	if err != nil {
+		return nil, fmt.Errorf("failed to write a subsetNode: %w", err)
+	}
+
+	return cidlink.Link{Cid: cid}, nil
+}
+
+func closeFile(bufferedWriter *bufio.Writer, currentFile *os.File) error {
+	err := bufferedWriter.Flush()
+	if err != nil {
+		return fmt.Errorf("failed to flush buffer: %w", err)
+	}
+
+	err = currentFile.Close()
+	if err != nil {
+		return fmt.Errorf("failed to close file: %w", err)
+	}
+	return nil
+}
+
+func writeNode(node datamodel.Node, w io.Writer) (cid.Cid, error) {
+	node = node.(schema.TypedNode).Representation()
+	var buf bytes.Buffer
+	err := dagcbor.Encode(node, &buf)
+	if err != nil {
+		return cid.Cid{}, err
+	}
+
+	data := buf.Bytes()
+
+	bd := cid.V1Builder{MhLength: -1, MhType: uint64(multicodec.Sha2_256), Codec: uint64(multicodec.DagCbor)}
+	cd, err := bd.Sum(data)
+	if err != nil {
+		return cid.Cid{}, err
+	}
+
+	c := cd.Bytes()
+
+	sizeVi := leb128.FromUInt64(uint64(len(c)) + uint64(len(data)))
+
+	if _, err := w.Write(sizeVi); err == nil {
+		if _, err := w.Write(c); err == nil {
+			if _, err := w.Write(data); err != nil {
+				return cid.Cid{}, err
+			}
+
+		}
+	}
+	return cd, nil
+}
diff --git a/main.go b/main.go
index e75a044b..8da32443 100644
--- a/main.go
+++ b/main.go
@@ -57,6 +57,7 @@ func main() {
 			newCmd_Version(),
 			newCmd_rpc(),
 			newCmd_check_deals(),
+			newCmd_SplitCar(),
 		},
 	}