Recordings
A recording is one take, and its files are the tracks: a video file, and on desktop and mobile a WAV master beside it. Every track carries its capture manifest, a record of what the recorder actually produced rather than what was asked for.
curl https://capturly.app/api/v1/recordings/{id} \
-H "Authorization: Bearer cap_live_..."Endpoints
| Method | Path |
|---|---|
GET | /v1/recordings — ready recordings, newest first; filter with session_id |
GET | /v1/recordings/{id} |
POST | /v1/recordings/{id}/files/{fileId}/download-url — a short-lived signed URL |
Reading needs recordings:read; minting a download URL needs
recordings:download.
A track
{
"id": "file_...",
"kind": "video",
"contentType": "video/mp4",
"sizeBytes": 734003200,
"durationSec": 1820,
"participantLabel": "Sam",
"consent": null,
"md5": "XrY7u+Ae7tCTyyK7j1rNww==",
"checksum": {
"algorithm": "md5",
"base64": "XrY7u+Ae7tCTyyK7j1rNww==",
"hex": "5eb63bbbe01eeed093cb22bb8f5acdc3",
"source": "object-store"
},
"capture": {
"audioSampleRate": 48000,
"audioBitDepth": null,
"audioChannels": 2,
"audioCodec": "mp4a.40.2",
"videoCodec": "avc1.640028",
"videoWidth": 1920,
"videoHeight": 1080,
"videoFrameRate": 30,
"framesDropped": 0,
"syncBeepAtMs": 1200,
"surface": "desktop",
"device": "Logitech BRIO",
"deviceType": "usb",
"appVersion": "2.15.0",
"capturedAt": "2026-11-13T12:00:00.000Z",
"watermarked": false
}
}The capture manifest
| Field | Meaning |
|---|---|
audioSampleRate, audioBitDepth, audioChannels, audioCodec | The audio in this file. On a WAV master they come from the file’s own header |
videoWidth, videoHeight, videoFrameRate, videoCodec | The picture in this file. Null on a WAV |
framesDropped | Video frames lost during the take. 0 is a clean take |
syncBeepAtMs | Offset of the 1 kHz alignment tone from the start of the file |
surface | What recorded it: web, desktop, ios, android or sdk |
device | The microphone (audio track) or camera (video track) as the operating system named it. Always null on capture-request tracks |
deviceType | What kind of input it was: built-in, wired, usb, bluetooth, virtual or unknown |
appVersion | The version of the app or SDK that recorded it |
capturedAt | When recording started, by the recorder’s clock |
watermarked | Whether the free-tier mark is in the frames |
durationSec sits on the track itself.
A value the recorder could not observe is null, never a plausible zero. Two
cases you will meet:
framesDroppedis null for Android takes, and for browsers without a camera drop counter (Firefox, Safari) unless the WebCodecs recorder counted it.- Tracks recorded by a client older than this field are null across the manifest.
The manifest is reported by the recorder. Treat it as provenance, not as a measurement Capturly re-derived from the file.
Capture-request tracks
A track recorded for a capture request is a
respondent’s, not yours, so it never carries the device’s name: an operating
system names devices after their owner (“Jane’s AirPods”). device is null
and deviceType says what kind it was. Capturly drops the name on the server
even if a client sends one.
These tracks also carry consent, the release the respondent accepted. It is
the same record GET /v1/capture-requests/{id}/submissions returns:
| Field | Meaning |
|---|---|
text | The release verbatim, as it was displayed. Editing the request later never changes it |
textSha256 | SHA-256 (hex) of text |
version | default:v1, custom:<sha>, or legacy for records backfilled at migration, which are not evidence of what was seen |
acceptedAt | When they accepted |
ipAddress, subjectName, subjectEmail | Captured at acceptance; email is null for open-link takes. Only returned to a key that also has capture_requests:read. For any other key these three fields are left out of the response, not set to null |
consent is null on every other track, and on a take whose request asked
for no release. Webhook payloads go to your own endpoint and
always carry the full record.
The checksum
checksum is never sent by the recorder. It is the MD5 of the stored bytes:
object-store: computed by storage as the upload landed.server: storage could not hash it (a WAV master assembled from a live backup, for one), so Capturly read the stored file back and hashed it. This happens within the hour; until thenchecksumis null.
MD5, not SHA-256, because storage computes it for free on every upload while
the bytes are already passing through. A SHA-256 would mean either reading
every file back after upload or trusting a hash the recording device sent. Use
it to confirm a download is intact: md5sum prints the hex form.