The error body

json

// A coded refusal: the sentence, the code to branch on,
// and the page that explains the code.
{ "detail": "separation takes exactly 1 audio file (got 2)",
  "code": "input_arity",
  "doc_url": "https://dueta.ai/docs/errors-and-limits#input_arity" }

// Not every refusal is coded. Those carry the sentence alone.
{ "detail": "Not found" }

// A JSON body that failed schema validation (422): one entry per field.
{ "detail": [ {
  "type": "value_error",
  "loc": ["body", "email"],
  "msg": "value is not a valid email address: An email address must have an @-sign."
} ] }

Status codes

Status	Meaning	Fix
400	The request reached the endpoint but could not be turned into an upload or a job: a bad filename, the wrong number of inputs, a malformed Content-Range, audio that will not decode, is under 0.5 s, or is digital silence, a total duration over the per-job cap, or Multi Mic Separation inputs that are not the same take.	Read detail; it names the field or the file. Do not retry unchanged.
401	No credential, a malformed one, a key that has been revoked or rotated away, a key sent to a console route that takes a signed-in session instead, a key of an account that was closed or disabled, or an organization key whose organization was closed or suspended.	Re-issue the key on the console's API keys page. A rotated key stops working the moment its replacement is shown. For a closed or disabled account, contact support: no request from it succeeds. Every 401 carries WWW-Authenticate: Bearer.
402	Not enough credit to cover the job, checked before it is queued. Uploads are never charged, so only job creation answers this. With an organization key, also spend_limit_reached: the organization or the key reached its monthly limit; context carries scope (org or key), limit_usd, used_usd and resets_at; limit_usd and used_usd are null for an organization's limit when the key's member may not see its billing.	Add credit in the console, then send the same request again: nothing was charged and the uploads are untouched. POST /v1/jobs/quote answers sufficient_credit before you commit. The context of insufficient_credit says where the shortfall is: organization, the name of the organization whose balance fell short, and billing_url, that balance's Billing page. organization is null for an account's own organization that has not been named. With an organization key and no billing permission, balance_usd, shortfall_usd and billing_url are null; the organization's Owners and Billing members add its credit and set its limits. A monthly limit resets at resets_at.
404	No upload, job, or stem with that name under your account; or a result that is not available yet or has been pruned.	Ids are scoped to the owning account. Take stem names from result.stems[].name and upload ids from the create response.
409	The request is out of order. Everything that stores or spends needs a confirmed email: upload create and content and job creation answer 409 email_unverified until it is. Upload content: the piece did not start at bytes_received, or the upload is already complete. Delete upload: a job still uses it. Job creation: an input is not ready, or a request under the same Idempotency-Key is still running. Cancel: the job is already terminal.	For an offset mismatch, detail names the byte to send from. For a cancel on a finished job, treat it as success. For an unverified email, click the link in the verification message. An input that is not ready carries the code input_not_ready and names the uploads that are not finished. Wait for each to complete, then resubmit the same ids. A key still in flight sends Retry-After; repeat the request with the same key after it.
410	The upload expired. An unused upload is swept with its bytes 24 hours after creation.	Create a new upload and send the file again. A job hardlinks its inputs, so expiry never takes a finished job's audio away.
413	The body exceeded the 100 MB per-file cap. A declared size_bytes over the cap is refused before a byte moves; a body that grows past it is refused mid-stream. Any other request body over 1 MiB, or a JSON job submission over 64 KiB, is refused before it is read, with no code.	Split or shorten the input. The upload is left failed; sending again from byte 0 reopens it. An oversized JSON body is a client bug: nothing the API reads is that large.
415	The file is video (.mp4, .mov, .mkv, .webm, .avi, .m4v by name, or a picture track found by the probe), or a PUT to content_url was sent as a multipart form.	Export the audio track (M4A, WAV, MP3) and upload that. To content_url, send the raw bytes with Content-Type: application/octet-stream.
422	A JSON body or query failed schema validation (a top-level field the endpoint does not read, a filter value outside its list, from after to on GET /v1/usage), ?format= on a stem download is not wav or flac, or an Idempotency-Key was reused for a different request body.	Schema validation: detail is an array of {loc, msg, type}; loc points at the offending field. A reused key: detail is a sentence with a code; branch on the code. A key names one request, so send a new key for a new one.
429	A rate limit, the ceiling on jobs in progress, the ceiling on recent unbilled jobs, or the ceiling on open event streams.	Wait Retry-After seconds, never zero. A rate limit is the one that carries retry_after_seconds and limit in the body; the two job ceilings do not, and clear when a job of yours ends rather than on a timer. Open event streams answer too_many_streams: close one, or poll.
503	A feature is closed rather than broken: FLAC transcoding is unavailable for a stem, or the job queue could not be reached. Or the service is busy: your balance was locked by another request for 5 s, or no connection was free ("The service is busy. Try again shortly.", no code). POST /v1/uploads/{id}/complete is busy the same way when no whole-file slot is free, or when your account already has 2 of its own uploads finishing on that server ("Your other uploads are still finishing").	For FLAC, download the WAV. For the queue and busy, wait Retry-After and send the same request; nothing was charged.
504	A FLAC transcode took too long.	Retry, or download the WAV.
5xx	A fault on our side.	Retry with exponential backoff. A failed job releases its hold and is never charged.

Rate limits

Route	Limit	Keyed on	Note
Every signed-in GET/HEAD	600 per minute	credential	All routes together. Per API key, or per signed-in account. RateLimit-* on the 429 only.
POST /v1/uploads	20 per minute	account	Its own budget. Submitting a job does not draw on it.
PUT /v1/uploads/{id}/content	20 per minute	account	Its own budget, every Content-Range piece counted. The same for POST /v1/uploads/{id}/complete.
POST /v1/jobs/music-separation	20 per minute	account	Its own budget. With source_url (a YouTube link) the same budget, plus the row below.
POST /v1/jobs/music-separation/probe	20 per minute	account	Its own budget. A probe is $0; its answer is good for 10 minutes and the same link is one probe.
Music Source Separation jobs fetched from a link	2 at once	account	Not a rate limit: queued or running jobs made with source_url. 429 with Retry-After until one finishes. Published by GET /v1/limits as products[music-separation].source_url.max_in_flight.
Links being read (probes)	2 at once	account	Not a rate limit: probes queued or running. 429 rate_limited with Retry-After. While 8 probes wait, a new one is 503 queue_full with Retry-After; a link already answered is never refused.
POST /v1/jobs/separation	20 per minute	account
POST /v1/jobs/multi-mic-separation	20 per minute	account	Counted apart from separation, not with it.
POST /v1/jobs/target-speaker-extraction	20 per minute	account	Its own budget.
POST /v1/jobs/noise-remover	20 per minute	account	Its own budget.
POST /v1/jobs/quote	600 per minute	account	Its own budget.
DELETE /v1/jobs/{id}	30 per minute	account	One budget with POST /v1/jobs/{id}/restore.
GET /v1/limits	60 per minute	IP	Cached 60 s; read it once, not per request.
Submitting a job by url	30 per hour	IP	Only the fetch-on-your-behalf path. Sending the file instead is unaffected.
GET /v1/jobs/{id}/events	30 per minute	account	The same for /v1/uploads/{id}/events. A browser session and an API key are counted apart.
Jobs in progress	1 at once, 4 after a purchase	account	Not a rate limit: queued or running. 429 too_many_active_jobs with Retry-After until one finishes.
Open event streams	16 at once	credential	Not a rate limit: concurrent SSE subscriptions. 429 too_many_streams with Retry-After. A browser session and an API key are counted apart.
Unbilled jobs	20 per hour	account	Not a rate limit: jobs that failed or were canceled without a charge.

Separate budgets. GET /v1/limits returns them, no credential.

Header	Meaning
RateLimit-Limit	The ceiling for the window.
RateLimit-Remaining	Requests left in it. 0 on the refusal itself.
RateLimit-Reset	Whole seconds until the window rolls.
X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset	The same three, older names.
Retry-After	Whole seconds to wait. On a 429 only.

http

HTTP/1.1 429 Too Many Requests
Retry-After: 60
RateLimit-Limit: 20
RateLimit-Remaining: 0
RateLimit-Reset: 60

{ "detail":
    "Too many requests. The limit here is 20 per minute; wait 60 seconds and try again.",
  "code": "rate_limited",
  "doc_url": "https://dueta.ai/docs/errors-and-limits#rate_limited",
  "retry_after_seconds": 60,
  "limit": "20 per minute" }

Five budgets answer 429. Branch on code.

code	Meaning
rate_limited	The request rate. Wait retry_after_seconds and repeat; RateLimit-* headers say where you stand.
too_many_active_jobs	Your jobs, not your requests: one at a time before the first purchase, four after. Retry when one of yours ends, not on a timer. No RateLimit-* headers.
too_many_in_flight	Your music link checks, not your requests. Retry when one ends, not on a timer. No RateLimit-* headers.
too_many_unbilled	Too many recent jobs finished without a charge. Clears as the window rolls.
too_many_streams	Open event streams on one credential. Close one, or poll the resource.
url_fetch_quota	The hourly budget for submitting a job by url. Send the file instead.

Honor Retry-After.

python

r = session.post(url, json=body, headers=headers)

if r.status_code == 429:
    code = r.json().get("code")
    if code == "rate_limited":
        time.sleep(int(r.headers["Retry-After"]))  # a window: waiting works
        r = session.post(url, json=body, headers=headers)
    elif code == "too_many_streams":
        ...                                        # an event stream: close one, or poll
    else:
        ...                                        # a job slot: wait for one to end

# Back off before the wall: a limited POST carries what is left (a read, on its 429 only).
if int(r.headers.get("RateLimit-Remaining", 1)) == 0:
    time.sleep(int(r.headers["RateLimit-Reset"]))

Codes

Coded refusals carry code and doc_url.

code	Status	Meaning
rate_limited	429	The request-rate limiter refused. Wait retry_after_seconds and repeat.
too_many_in_flight	429	This account already has as many music link checks or link-fetched jobs running as it may. Retry when one of them ends, not on a timer.
too_many_active_jobs	429	This account already has context.limit jobs queued or running: one before its first purchase, four after (context.tier). Retry when one ends, or cancel one.
too_many_unbilled	429	Too many recent jobs ended without a charge. Clears as the window rolls.
too_many_streams	429	Close one of this credential's open event streams, or poll the resource instead. context.max_open_streams is the cap per server process.
url_fetch_quota	429	Send the file instead: the per-address budget for fetching a submitted URL is spent.
queue_full	503	The shared queue is full. Retry after retry_after_seconds; the context carries the depth and the estimated wait.
gpu_unavailable	503	The worker is alive and has declined to take work: its GPU breaker is open. Retry after retry_after_seconds.
input_arity	400	The product does not take this many files.
input_not_ready	409	One or more named uploads has not finished transferring.
upload_failed	409	One or more named uploads failed (its error says why: unreadable, silent, too long). Sending the same bytes again fails the same way: fix the file and create a new upload. GET /v1/uploads/{id} error_code stays upload_failed; its reason_code is the specific input_* refusal (input_unreadable, input_too_short, input_silent, input_too_long, input_audio_format, input_is_video, input_file_too_large), or null.
input_total_too_long	400	Every input added together is over the per-job ceiling.
input_too_long	400	One recording is longer than the step processes in a single pass.
input_range_invalid	422	range cannot be used as sent: the recording is 30 s or shorter (a range is offered only on longer ones), start_seconds is not before end_seconds, the range does not lie within the recording (for a source_url, the length its probe read), or it is under 30 s. context carries start_seconds, end_seconds, duration_seconds and min_seconds.
submission_collided	409	The same submission was sent twice at once and the other one finished before this one could be matched to it. Nothing was charged. Send it again after Retry-After; context carries retry_after_seconds.
input_range_busy	503	range could not be cut right now because the service is busy. Nothing was charged and no job was made. Send the same request again after Retry-After; context carries retry_after_seconds.
input_not_one_take	400	The set is not one take: the longest is too many times the shortest.
input_unequal_length	400	The model needs every input to be the same length, and they are not.
input_too_many_files	400	More input files than one job may carry.
input_file_too_large	413	One file is over the per-file byte ceiling.
upload_quota	429	The account holds as many unused uploads as it may (20 still being sent, 100 in all, or 5 GB), or sent 50 GB in the last 24 hours. Use, finish or delete an upload, or wait. Unused uploads expire after a day.
input_unreadable	400	The file is not audio this service can read: empty, damaged, non-finite samples, or not an audio format at all. Export it again as audio.
input_too_short	400, failed	The recording is shorter than the 0.5 s minimum. Target Speaker Extraction also fails a recording under 2 s, with details seconds and min_seconds; the job was not charged.
input_silent	400	The recording holds only digital silence.
input_audio_format	400	The audio has more than 8 channels or a sample rate above 192 kHz.
input_is_video	415	The file carries a video track. Export the audio track and send that.
input_file_type	400	The file's extension is not an audio format this service reads; context.allowed lists the ones it does.
input_filename	400	The filename is empty, longer than 200 bytes (66 Korean characters), or contains control characters. When it is too long, context carries max_bytes and max_korean_chars.
insufficient_credit	402	Add credit: the quote is more than the balance holds. context carries balance_usd, required_usd and the shortfall_usd to cover; balance_usd and shortfall_usd are null for an organization key without the billing permission.
spend_limit_reached	402	This job would take an organization, or one of its API keys, past its monthly limit. context carries the job's required_usd and, when you may see them, limit_usd and used_usd.
email_unverified	409	The address has never been confirmed. Confirm it from the link in the signup mail, or ask for a new one from Settings; nothing else unblocks the request.
download_limit_reached	403	This job's archive cannot be downloaded again. Single stems, previews and the original uploads are unaffected.
purchase_required	403	Returned only when a purchase gate is enabled. Downloads currently do not require a credit purchase.
phone_invalid	422	The number is not one an SMS can reach: write it in E.164 (+821012345678) or, for a Korean mobile, as the country writes it (010-1234-5678).
sms_unavailable	503	No code could be sent or checked: SMS is not configured on this deployment, or the provider did not answer. When the provider could not be reached nothing was consumed; when it did not answer in time the send or try counts, since the SMS may have gone out. Retry-After and context.retry_after_seconds say when to try again.
phone_send_cooldown	429	A code was sent too recently, or the day's sends for this account or number are spent; context.retry_after_seconds says when the next may go.
phone_code_invalid	400	The code did not match, expired, or was voided after too many wrong tries. Ask for a new one; the number is unchanged.
phone_in_use	409	The number is already verified on another account. One account per number.
invalid_api_key	401	The key is not a live key: mistyped, revoked, rotated away, or its organization was closed. Copy it again, or create a new one in the console.
api_key_placement	401	Send the API key as Authorization: Bearer <key>, not in another header.
session_only	401	This route takes a signed-in session, not an API key. A key opens the routes on these docs.
coupon_invalid · Console only	404	The code does not exist, is switched off, or is not eight to sixteen letters and digits. Check the spelling.
coupon_expired · Console only	410	The code's expiry date has passed. Nothing was applied.
coupon_not_started · Console only	422	The code is not active yet; context.starts_at says when it opens.
coupon_exhausted · Console only	410	Every use of this code has been taken. Nothing was applied.
coupon_used · Console only	409	This account has already used the code as many times as it allows. Nothing was applied.
coupon_kind · Console only	422	A credit code was sent to the checkout, or a discount code to redeem: use the other route.
callback_already_registered	409	The job already has a different callback_url, and a callback cannot be moved once registered. Read the job to see where its deliveries go; context carries job_id and the callback_url that stands. The same URL twice is an ordinary retry and is not this.
idempotency_in_flight	409	Wait and retry: the first request under this key has not answered yet, and the retry gets its answer. Retry-After says how long.
idempotency_body_mismatch	422	Use a new key. This one already names a different request, and no amount of waiting changes that.
upload_in_progress	409	Wait and resume: another transfer holds this upload's writer lease and is still sending.
upload_lease_lost	409	Create a new upload and send the file again: this transfer's lease went to another one, so nothing it sent was kept.
upload_already_ready	409	Nothing to send: this upload already has every byte and is certified. Create a new upload for a different recording.
upload_offset_mismatch	409	Resume from context.bytes_received, which is where the file actually ends; context.sent_from is where this piece started.
upload_not_found	404	No upload by that id. Check the ids in context.upload_ids, then create one with POST /v1/uploads and send its bytes before submitting.
body_shape	400	The body is not a submission. Send inputs naming uploads, or url, or the file as multipart: one of the three.
currency_unsupported	422	Leave currency out, or send one of context.supported: no payment account here charges context.currency.
enrollment_is_mixture	422	Send a separate sample of the target speaker as inputs[1]: it is the same audio as the recording in inputs[0].
enrollment_too_short	422, failed	Send a longer sample of the target speaker: 3 s of speech at least, 10 s or more recommended. A refusal's context carries min_seconds and given_seconds; a failed job was not charged.
enrollment_too_long	422	Send a sample of the target speaker of 300 s or less, after the recording. context carries max_seconds and given_seconds.
enrollment_no_speech	failed	Send a sample in which the target speaker talks. The job failed and was not charged.
silent_output	failed	Output too quiet. The job failed and was not charged.
target_not_found	failed	Check the sample is the voice of someone in the recording. The job failed and was not charged.
input_too_many_channels	422	Send a mono or stereo file: Music Source Separation takes nothing wider. context carries channels and max_channels.
music_too_many_channels	failed	Send a mono or stereo file: this one has more than two channels (context.channels). The job failed and was not charged.
probe_invalid	422	probe_id cannot price this link: not your probe or older than 10 minutes (context.state unknown), not answered yet (pending), or a probe of another video (mismatch). Probe the link again.
source_url_invalid	422	Send a YouTube video link (youtube.com/watch?v=..., youtu.be/..., a short): this is not one, or it names a playlist, a channel or another site. context carries url.
source_unavailable	failed	The video cannot be fetched: private, removed, restricted in this region or by age, or YouTube refused the request. The job failed and was not charged.
source_blocked	failed	The platform refused our server's request (a bot check, a 403 or a 429): a fault on our side, not the video's state (that is source_unavailable). Try again later. The job failed and was not charged.
source_too_long	failed	The video is longer than the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
source_live	failed	The link is a live or upcoming stream; only a finished video can be separated. The job failed and was not charged.
source_fetch_failed	failed	Fetching the video did not finish (a timeout, the size limit, a decode failure). Retry the job; it was not charged.
source_insufficient_credit	failed	The song is longer than the credit held for it could pay for: the hold was the smaller of the length cap's cost and your balance. Top up and submit the link again. details carries seconds, cost_usd and held_usd. The job failed and was not charged.
music_too_long	failed	Send a song within the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
input_malware_detected	failed	Upload a clean copy of the file: the malware scan flagged this input and it was not processed. The job failed and was not charged.
input_quiet	400	The whole recording sits far below speech level.
input_mostly_silent	400	Almost none of the recording is above the silence floor.
input_narrowband	400	The source was captured below the rate the model works at.
preview_unavailable	404	Play and draw from the WAV stems instead: this job has no preview or peaks (it finished before they existed, or they could not be made). result.previews says so before you ask.
not_allowed	403	The caller is authenticated but may not do this, and the route named no more specific code. Check the key or session belongs to the account or organization that owns the resource.
not_found	404	No such resource or path, or one this caller may not see, and the route named no more specific code. Check the id and the path against the endpoint index.
method_not_allowed	405	The path exists and does not take this HTTP method. Check the method against the endpoint index.

A failed job carries error_code from this list, or null.

error_code	Meaning
enrollment_no_speech	Send a sample in which the target speaker talks. The job failed and was not charged.
enrollment_too_short	Send a longer sample of the target speaker: 3 s of speech at least, 10 s or more recommended. A refusal's context carries min_seconds and given_seconds; a failed job was not charged.
target_not_found	Check the sample is the voice of someone in the recording. The job failed and was not charged.
input_too_short	The recording is shorter than the 0.5 s minimum. Target Speaker Extraction also fails a recording under 2 s, with details seconds and min_seconds; the job was not charged.
music_too_many_channels	Send a mono or stereo file: this one has more than two channels (context.channels). The job failed and was not charged.
music_too_long	Send a song within the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
source_unavailable	The video cannot be fetched: private, removed, restricted in this region or by age, or YouTube refused the request. The job failed and was not charged.
source_too_long	The video is longer than the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
source_live	The link is a live or upcoming stream; only a finished video can be separated. The job failed and was not charged.
source_fetch_failed	Fetching the video did not finish (a timeout, the size limit, a decode failure). Retry the job; it was not charged.
source_insufficient_credit	The song is longer than the credit held for it could pay for: the hold was the smaller of the length cap's cost and your balance. Top up and submit the link again. details carries seconds, cost_usd and held_usd. The job failed and was not charged.
silent_output	Output too quiet. The job failed and was not charged.
source_blocked	The platform refused our server's request (a bot check, a 403 or a 429): a fault on our side, not the video's state (that is source_unavailable). Try again later. The job failed and was not charged.
input_malware_detected	Upload a clean copy of the file: the malware scan flagged this input and it was not processed. The job failed and was not charged.

Limits & retention

Limit	Value
Formats	WAV, AIFF, FLAC, MP3, M4A / AAC, OGG / Opus, WMA, CAF, or AMR. Video refused.
File size	100 MB per file.
30 min under 100 MB	MP3 or M4A.
Duration per file	At least 0.5 s, not digital silence.
Duplex Conversation Separation recording	30 min.
Duration per job	120 min, all inputs summed.
Inputs per job	Duplex Conversation Separation 1. Multi Mic Separation 2–10, longest at most 2× the shortest.
Unused upload	Swept after 24 hours.
Results and inputs	Kept 7 days after the job finishes; after a purchase (demo and full refunds excluded), until you delete them. An organization key's jobs follow the organization's purchases.
API keys	Do not expire.

API reference

The error body

json

// A coded refusal: the sentence, the code to branch on,
// and the page that explains the code.
{ "detail": "separation takes exactly 1 audio file (got 2)",
  "code": "input_arity",
  "doc_url": "https://dueta.ai/docs/errors-and-limits#input_arity" }

// Not every refusal is coded. Those carry the sentence alone.
{ "detail": "Not found" }

// A JSON body that failed schema validation (422): one entry per field.
{ "detail": [ {
  "type": "value_error",
  "loc": ["body", "email"],
  "msg": "value is not a valid email address: An email address must have an @-sign."
} ] }

Status codes

Status	Meaning	Fix
400	The request reached the endpoint but could not be turned into an upload or a job: a bad filename, the wrong number of inputs, a malformed Content-Range, audio that will not decode, is under 0.5 s, or is digital silence, a total duration over the per-job cap, or Multi Mic Separation inputs that are not the same take.	Read detail; it names the field or the file. Do not retry unchanged.
401	No credential, a malformed one, a key that has been revoked or rotated away, a key sent to a console route that takes a signed-in session instead, a key of an account that was closed or disabled, or an organization key whose organization was closed or suspended.	Re-issue the key on the console's API keys page. A rotated key stops working the moment its replacement is shown. For a closed or disabled account, contact support: no request from it succeeds. Every 401 carries WWW-Authenticate: Bearer.
402	Not enough credit to cover the job, checked before it is queued. Uploads are never charged, so only job creation answers this. With an organization key, also spend_limit_reached: the organization or the key reached its monthly limit; context carries scope (org or key), limit_usd, used_usd and resets_at; limit_usd and used_usd are null for an organization's limit when the key's member may not see its billing.	Add credit in the console, then send the same request again: nothing was charged and the uploads are untouched. POST /v1/jobs/quote answers sufficient_credit before you commit. The context of insufficient_credit says where the shortfall is: organization, the name of the organization whose balance fell short, and billing_url, that balance's Billing page. organization is null for an account's own organization that has not been named. With an organization key and no billing permission, balance_usd, shortfall_usd and billing_url are null; the organization's Owners and Billing members add its credit and set its limits. A monthly limit resets at resets_at.
404	No upload, job, or stem with that name under your account; or a result that is not available yet or has been pruned.	Ids are scoped to the owning account. Take stem names from result.stems[].name and upload ids from the create response.
409	The request is out of order. Everything that stores or spends needs a confirmed email: upload create and content and job creation answer 409 email_unverified until it is. Upload content: the piece did not start at bytes_received, or the upload is already complete. Delete upload: a job still uses it. Job creation: an input is not ready, or a request under the same Idempotency-Key is still running. Cancel: the job is already terminal.	For an offset mismatch, detail names the byte to send from. For a cancel on a finished job, treat it as success. For an unverified email, click the link in the verification message. An input that is not ready carries the code input_not_ready and names the uploads that are not finished. Wait for each to complete, then resubmit the same ids. A key still in flight sends Retry-After; repeat the request with the same key after it.
410	The upload expired. An unused upload is swept with its bytes 24 hours after creation.	Create a new upload and send the file again. A job hardlinks its inputs, so expiry never takes a finished job's audio away.
413	The body exceeded the 100 MB per-file cap. A declared size_bytes over the cap is refused before a byte moves; a body that grows past it is refused mid-stream. Any other request body over 1 MiB, or a JSON job submission over 64 KiB, is refused before it is read, with no code.	Split or shorten the input. The upload is left failed; sending again from byte 0 reopens it. An oversized JSON body is a client bug: nothing the API reads is that large.
415	The file is video (.mp4, .mov, .mkv, .webm, .avi, .m4v by name, or a picture track found by the probe), or a PUT to content_url was sent as a multipart form.	Export the audio track (M4A, WAV, MP3) and upload that. To content_url, send the raw bytes with Content-Type: application/octet-stream.
422	A JSON body or query failed schema validation (a top-level field the endpoint does not read, a filter value outside its list, from after to on GET /v1/usage), ?format= on a stem download is not wav or flac, or an Idempotency-Key was reused for a different request body.	Schema validation: detail is an array of {loc, msg, type}; loc points at the offending field. A reused key: detail is a sentence with a code; branch on the code. A key names one request, so send a new key for a new one.
429	A rate limit, the ceiling on jobs in progress, the ceiling on recent unbilled jobs, or the ceiling on open event streams.	Wait Retry-After seconds, never zero. A rate limit is the one that carries retry_after_seconds and limit in the body; the two job ceilings do not, and clear when a job of yours ends rather than on a timer. Open event streams answer too_many_streams: close one, or poll.
503	A feature is closed rather than broken: FLAC transcoding is unavailable for a stem, or the job queue could not be reached. Or the service is busy: your balance was locked by another request for 5 s, or no connection was free ("The service is busy. Try again shortly.", no code). POST /v1/uploads/{id}/complete is busy the same way when no whole-file slot is free, or when your account already has 2 of its own uploads finishing on that server ("Your other uploads are still finishing").	For FLAC, download the WAV. For the queue and busy, wait Retry-After and send the same request; nothing was charged.
504	A FLAC transcode took too long.	Retry, or download the WAV.
5xx	A fault on our side.	Retry with exponential backoff. A failed job releases its hold and is never charged.

Rate limits

Route	Limit	Keyed on	Note
Every signed-in GET/HEAD	600 per minute	credential	All routes together. Per API key, or per signed-in account. RateLimit-* on the 429 only.
POST /v1/uploads	20 per minute	account	Its own budget. Submitting a job does not draw on it.
PUT /v1/uploads/{id}/content	20 per minute	account	Its own budget, every Content-Range piece counted. The same for POST /v1/uploads/{id}/complete.
POST /v1/jobs/music-separation	20 per minute	account	Its own budget. With source_url (a YouTube link) the same budget, plus the row below.
POST /v1/jobs/music-separation/probe	20 per minute	account	Its own budget. A probe is $0; its answer is good for 10 minutes and the same link is one probe.
Music Source Separation jobs fetched from a link	2 at once	account	Not a rate limit: queued or running jobs made with source_url. 429 with Retry-After until one finishes. Published by GET /v1/limits as products[music-separation].source_url.max_in_flight.
Links being read (probes)	2 at once	account	Not a rate limit: probes queued or running. 429 rate_limited with Retry-After. While 8 probes wait, a new one is 503 queue_full with Retry-After; a link already answered is never refused.
POST /v1/jobs/separation	20 per minute	account
POST /v1/jobs/multi-mic-separation	20 per minute	account	Counted apart from separation, not with it.
POST /v1/jobs/target-speaker-extraction	20 per minute	account	Its own budget.
POST /v1/jobs/noise-remover	20 per minute	account	Its own budget.
POST /v1/jobs/quote	600 per minute	account	Its own budget.
DELETE /v1/jobs/{id}	30 per minute	account	One budget with POST /v1/jobs/{id}/restore.
GET /v1/limits	60 per minute	IP	Cached 60 s; read it once, not per request.
Submitting a job by url	30 per hour	IP	Only the fetch-on-your-behalf path. Sending the file instead is unaffected.
GET /v1/jobs/{id}/events	30 per minute	account	The same for /v1/uploads/{id}/events. A browser session and an API key are counted apart.
Jobs in progress	1 at once, 4 after a purchase	account	Not a rate limit: queued or running. 429 too_many_active_jobs with Retry-After until one finishes.
Open event streams	16 at once	credential	Not a rate limit: concurrent SSE subscriptions. 429 too_many_streams with Retry-After. A browser session and an API key are counted apart.
Unbilled jobs	20 per hour	account	Not a rate limit: jobs that failed or were canceled without a charge.

Separate budgets. GET /v1/limits returns them, no credential.

Header	Meaning
RateLimit-Limit	The ceiling for the window.
RateLimit-Remaining	Requests left in it. 0 on the refusal itself.
RateLimit-Reset	Whole seconds until the window rolls.
X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset	The same three, older names.
Retry-After	Whole seconds to wait. On a 429 only.

http

HTTP/1.1 429 Too Many Requests
Retry-After: 60
RateLimit-Limit: 20
RateLimit-Remaining: 0
RateLimit-Reset: 60

{ "detail":
    "Too many requests. The limit here is 20 per minute; wait 60 seconds and try again.",
  "code": "rate_limited",
  "doc_url": "https://dueta.ai/docs/errors-and-limits#rate_limited",
  "retry_after_seconds": 60,
  "limit": "20 per minute" }

Five budgets answer 429. Branch on code.

code	Meaning
rate_limited	The request rate. Wait retry_after_seconds and repeat; RateLimit-* headers say where you stand.
too_many_active_jobs	Your jobs, not your requests: one at a time before the first purchase, four after. Retry when one of yours ends, not on a timer. No RateLimit-* headers.
too_many_in_flight	Your music link checks, not your requests. Retry when one ends, not on a timer. No RateLimit-* headers.
too_many_unbilled	Too many recent jobs finished without a charge. Clears as the window rolls.
too_many_streams	Open event streams on one credential. Close one, or poll the resource.
url_fetch_quota	The hourly budget for submitting a job by url. Send the file instead.

Honor Retry-After.

python

r = session.post(url, json=body, headers=headers)

if r.status_code == 429:
    code = r.json().get("code")
    if code == "rate_limited":
        time.sleep(int(r.headers["Retry-After"]))  # a window: waiting works
        r = session.post(url, json=body, headers=headers)
    elif code == "too_many_streams":
        ...                                        # an event stream: close one, or poll
    else:
        ...                                        # a job slot: wait for one to end

# Back off before the wall: a limited POST carries what is left (a read, on its 429 only).
if int(r.headers.get("RateLimit-Remaining", 1)) == 0:
    time.sleep(int(r.headers["RateLimit-Reset"]))

Codes

Coded refusals carry code and doc_url.

code	Status	Meaning
rate_limited	429	The request-rate limiter refused. Wait retry_after_seconds and repeat.
too_many_in_flight	429	This account already has as many music link checks or link-fetched jobs running as it may. Retry when one of them ends, not on a timer.
too_many_active_jobs	429	This account already has context.limit jobs queued or running: one before its first purchase, four after (context.tier). Retry when one ends, or cancel one.
too_many_unbilled	429	Too many recent jobs ended without a charge. Clears as the window rolls.
too_many_streams	429	Close one of this credential's open event streams, or poll the resource instead. context.max_open_streams is the cap per server process.
url_fetch_quota	429	Send the file instead: the per-address budget for fetching a submitted URL is spent.
queue_full	503	The shared queue is full. Retry after retry_after_seconds; the context carries the depth and the estimated wait.
gpu_unavailable	503	The worker is alive and has declined to take work: its GPU breaker is open. Retry after retry_after_seconds.
input_arity	400	The product does not take this many files.
input_not_ready	409	One or more named uploads has not finished transferring.
upload_failed	409	One or more named uploads failed (its error says why: unreadable, silent, too long). Sending the same bytes again fails the same way: fix the file and create a new upload. GET /v1/uploads/{id} error_code stays upload_failed; its reason_code is the specific input_* refusal (input_unreadable, input_too_short, input_silent, input_too_long, input_audio_format, input_is_video, input_file_too_large), or null.
input_total_too_long	400	Every input added together is over the per-job ceiling.
input_too_long	400	One recording is longer than the step processes in a single pass.
input_range_invalid	422	range cannot be used as sent: the recording is 30 s or shorter (a range is offered only on longer ones), start_seconds is not before end_seconds, the range does not lie within the recording (for a source_url, the length its probe read), or it is under 30 s. context carries start_seconds, end_seconds, duration_seconds and min_seconds.
submission_collided	409	The same submission was sent twice at once and the other one finished before this one could be matched to it. Nothing was charged. Send it again after Retry-After; context carries retry_after_seconds.
input_range_busy	503	range could not be cut right now because the service is busy. Nothing was charged and no job was made. Send the same request again after Retry-After; context carries retry_after_seconds.
input_not_one_take	400	The set is not one take: the longest is too many times the shortest.
input_unequal_length	400	The model needs every input to be the same length, and they are not.
input_too_many_files	400	More input files than one job may carry.
input_file_too_large	413	One file is over the per-file byte ceiling.
upload_quota	429	The account holds as many unused uploads as it may (20 still being sent, 100 in all, or 5 GB), or sent 50 GB in the last 24 hours. Use, finish or delete an upload, or wait. Unused uploads expire after a day.
input_unreadable	400	The file is not audio this service can read: empty, damaged, non-finite samples, or not an audio format at all. Export it again as audio.
input_too_short	400, failed	The recording is shorter than the 0.5 s minimum. Target Speaker Extraction also fails a recording under 2 s, with details seconds and min_seconds; the job was not charged.
input_silent	400	The recording holds only digital silence.
input_audio_format	400	The audio has more than 8 channels or a sample rate above 192 kHz.
input_is_video	415	The file carries a video track. Export the audio track and send that.
input_file_type	400	The file's extension is not an audio format this service reads; context.allowed lists the ones it does.
input_filename	400	The filename is empty, longer than 200 bytes (66 Korean characters), or contains control characters. When it is too long, context carries max_bytes and max_korean_chars.
insufficient_credit	402	Add credit: the quote is more than the balance holds. context carries balance_usd, required_usd and the shortfall_usd to cover; balance_usd and shortfall_usd are null for an organization key without the billing permission.
spend_limit_reached	402	This job would take an organization, or one of its API keys, past its monthly limit. context carries the job's required_usd and, when you may see them, limit_usd and used_usd.
email_unverified	409	The address has never been confirmed. Confirm it from the link in the signup mail, or ask for a new one from Settings; nothing else unblocks the request.
download_limit_reached	403	This job's archive cannot be downloaded again. Single stems, previews and the original uploads are unaffected.
purchase_required	403	Returned only when a purchase gate is enabled. Downloads currently do not require a credit purchase.
phone_invalid	422	The number is not one an SMS can reach: write it in E.164 (+821012345678) or, for a Korean mobile, as the country writes it (010-1234-5678).
sms_unavailable	503	No code could be sent or checked: SMS is not configured on this deployment, or the provider did not answer. When the provider could not be reached nothing was consumed; when it did not answer in time the send or try counts, since the SMS may have gone out. Retry-After and context.retry_after_seconds say when to try again.
phone_send_cooldown	429	A code was sent too recently, or the day's sends for this account or number are spent; context.retry_after_seconds says when the next may go.
phone_code_invalid	400	The code did not match, expired, or was voided after too many wrong tries. Ask for a new one; the number is unchanged.
phone_in_use	409	The number is already verified on another account. One account per number.
invalid_api_key	401	The key is not a live key: mistyped, revoked, rotated away, or its organization was closed. Copy it again, or create a new one in the console.
api_key_placement	401	Send the API key as Authorization: Bearer <key>, not in another header.
session_only	401	This route takes a signed-in session, not an API key. A key opens the routes on these docs.
coupon_invalid · Console only	404	The code does not exist, is switched off, or is not eight to sixteen letters and digits. Check the spelling.
coupon_expired · Console only	410	The code's expiry date has passed. Nothing was applied.
coupon_not_started · Console only	422	The code is not active yet; context.starts_at says when it opens.
coupon_exhausted · Console only	410	Every use of this code has been taken. Nothing was applied.
coupon_used · Console only	409	This account has already used the code as many times as it allows. Nothing was applied.
coupon_kind · Console only	422	A credit code was sent to the checkout, or a discount code to redeem: use the other route.
callback_already_registered	409	The job already has a different callback_url, and a callback cannot be moved once registered. Read the job to see where its deliveries go; context carries job_id and the callback_url that stands. The same URL twice is an ordinary retry and is not this.
idempotency_in_flight	409	Wait and retry: the first request under this key has not answered yet, and the retry gets its answer. Retry-After says how long.
idempotency_body_mismatch	422	Use a new key. This one already names a different request, and no amount of waiting changes that.
upload_in_progress	409	Wait and resume: another transfer holds this upload's writer lease and is still sending.
upload_lease_lost	409	Create a new upload and send the file again: this transfer's lease went to another one, so nothing it sent was kept.
upload_already_ready	409	Nothing to send: this upload already has every byte and is certified. Create a new upload for a different recording.
upload_offset_mismatch	409	Resume from context.bytes_received, which is where the file actually ends; context.sent_from is where this piece started.
upload_not_found	404	No upload by that id. Check the ids in context.upload_ids, then create one with POST /v1/uploads and send its bytes before submitting.
body_shape	400	The body is not a submission. Send inputs naming uploads, or url, or the file as multipart: one of the three.
currency_unsupported	422	Leave currency out, or send one of context.supported: no payment account here charges context.currency.
enrollment_is_mixture	422	Send a separate sample of the target speaker as inputs[1]: it is the same audio as the recording in inputs[0].
enrollment_too_short	422, failed	Send a longer sample of the target speaker: 3 s of speech at least, 10 s or more recommended. A refusal's context carries min_seconds and given_seconds; a failed job was not charged.
enrollment_too_long	422	Send a sample of the target speaker of 300 s or less, after the recording. context carries max_seconds and given_seconds.
enrollment_no_speech	failed	Send a sample in which the target speaker talks. The job failed and was not charged.
silent_output	failed	Output too quiet. The job failed and was not charged.
target_not_found	failed	Check the sample is the voice of someone in the recording. The job failed and was not charged.
input_too_many_channels	422	Send a mono or stereo file: Music Source Separation takes nothing wider. context carries channels and max_channels.
music_too_many_channels	failed	Send a mono or stereo file: this one has more than two channels (context.channels). The job failed and was not charged.
probe_invalid	422	probe_id cannot price this link: not your probe or older than 10 minutes (context.state unknown), not answered yet (pending), or a probe of another video (mismatch). Probe the link again.
source_url_invalid	422	Send a YouTube video link (youtube.com/watch?v=..., youtu.be/..., a short): this is not one, or it names a playlist, a channel or another site. context carries url.
source_unavailable	failed	The video cannot be fetched: private, removed, restricted in this region or by age, or YouTube refused the request. The job failed and was not charged.
source_blocked	failed	The platform refused our server's request (a bot check, a 403 or a 429): a fault on our side, not the video's state (that is source_unavailable). Try again later. The job failed and was not charged.
source_too_long	failed	The video is longer than the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
source_live	failed	The link is a live or upcoming stream; only a finished video can be separated. The job failed and was not charged.
source_fetch_failed	failed	Fetching the video did not finish (a timeout, the size limit, a decode failure). Retry the job; it was not charged.
source_insufficient_credit	failed	The song is longer than the credit held for it could pay for: the hold was the smaller of the length cap's cost and your balance. Top up and submit the link again. details carries seconds, cost_usd and held_usd. The job failed and was not charged.
music_too_long	failed	Send a song within the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
input_malware_detected	failed	Upload a clean copy of the file: the malware scan flagged this input and it was not processed. The job failed and was not charged.
input_quiet	400	The whole recording sits far below speech level.
input_mostly_silent	400	Almost none of the recording is above the silence floor.
input_narrowband	400	The source was captured below the rate the model works at.
preview_unavailable	404	Play and draw from the WAV stems instead: this job has no preview or peaks (it finished before they existed, or they could not be made). result.previews says so before you ask.
not_allowed	403	The caller is authenticated but may not do this, and the route named no more specific code. Check the key or session belongs to the account or organization that owns the resource.
not_found	404	No such resource or path, or one this caller may not see, and the route named no more specific code. Check the id and the path against the endpoint index.
method_not_allowed	405	The path exists and does not take this HTTP method. Check the method against the endpoint index.

A failed job carries error_code from this list, or null.

error_code	Meaning
enrollment_no_speech	Send a sample in which the target speaker talks. The job failed and was not charged.
enrollment_too_short	Send a longer sample of the target speaker: 3 s of speech at least, 10 s or more recommended. A refusal's context carries min_seconds and given_seconds; a failed job was not charged.
target_not_found	Check the sample is the voice of someone in the recording. The job failed and was not charged.
input_too_short	The recording is shorter than the 0.5 s minimum. Target Speaker Extraction also fails a recording under 2 s, with details seconds and min_seconds; the job was not charged.
music_too_many_channels	Send a mono or stereo file: this one has more than two channels (context.channels). The job failed and was not charged.
music_too_long	Send a song within the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
source_unavailable	The video cannot be fetched: private, removed, restricted in this region or by age, or YouTube refused the request. The job failed and was not charged.
source_too_long	The video is longer than the length limit (GET /v1/limits, music-separation max_input_seconds); context carries seconds and max_seconds. The job failed and was not charged.
source_live	The link is a live or upcoming stream; only a finished video can be separated. The job failed and was not charged.
source_fetch_failed	Fetching the video did not finish (a timeout, the size limit, a decode failure). Retry the job; it was not charged.
source_insufficient_credit	The song is longer than the credit held for it could pay for: the hold was the smaller of the length cap's cost and your balance. Top up and submit the link again. details carries seconds, cost_usd and held_usd. The job failed and was not charged.
silent_output	Output too quiet. The job failed and was not charged.
source_blocked	The platform refused our server's request (a bot check, a 403 or a 429): a fault on our side, not the video's state (that is source_unavailable). Try again later. The job failed and was not charged.
input_malware_detected	Upload a clean copy of the file: the malware scan flagged this input and it was not processed. The job failed and was not charged.

Limits & retention

Limit	Value
Formats	WAV, AIFF, FLAC, MP3, M4A / AAC, OGG / Opus, WMA, CAF, or AMR. Video refused.
File size	100 MB per file.
30 min under 100 MB	MP3 or M4A.
Duration per file	At least 0.5 s, not digital silence.
Duplex Conversation Separation recording	30 min.
Duration per job	120 min, all inputs summed.
Inputs per job	Duplex Conversation Separation 1. Multi Mic Separation 2–10, longest at most 2× the shortest.
Unused upload	Swept after 24 hours.
Results and inputs	Kept 7 days after the job finishes; after a purchase (demo and full refunds excluded), until you delete them. An organization key's jobs follow the organization's purchases.
API keys	Do not expire.

API reference

Errors & limits

The error body

Status codes

Rate limits

Codes

Limits & retention

Errors & limits

The error body

Status codes

Rate limits

Codes

Limits & retention