Send files directly between devices over the internet - no cloud storage, no size limits.
@altersend/drive
Chunked file transfer over a caller-supplied channel. The sender reads byte ranges from the source, the receiver writes them to offsets in the destination. Neither side makes an intermediate copy.
Usage
import { sendFile, receiveFile } from '@altersend/drive'
const savedTo = await receiveFile(destPath, channelB)
await sendFile(srcPath, channelA)
Drive(receiveDir) wraps these when one directory is reused:
drive.receive(name, channel) saves under it, drive.send(path, channel) sends
any file on disk.
transferId is optional — the sender generates one, the receiver adopts it. Pass
one when a channel carries several transfers.
For a non-disk source, pass your own reader/writer to SenderSession /
ReceiverSession.
Protocol
start { transferId, name, size, chunkSize } sender → receiver
need { indices } receiver → sender
<chunk> header{ index } + bytes sender → receiver
complete { } sender → receiver
ack { savedTo } receiver → sender
cancel { reason? } either
Chunk size is derived from file size, so both sides compute the same geometry
from start. The engine verifies geometry only — it does not hash content, so a
receiver cannot detect a peer serving the wrong bytes. That is fine while the
only source is the authenticated sender; a seeder would need a commitment the
receiver can check independently.
Interfaces
The engine imports no filesystem or socket:
| Interface | Native | Browser (not built) |
|---|---|---|
ChunkReader |
pread | File.slice().arrayBuffer() |
ChunkWriter |
pwrite | File System Access API / OPFS |
DriveChannel |
Protomux on UDX | WebSocket |
Ships DiskReader / DiskWriter. They import #fs and #path, which the
imports map resolves to bare-fs/promises / bare-path under Bare and the Node
builtins elsewhere, so the same adapter runs in the worklet and in tests. The
root export resolves to an fs-free build under a browser condition.