Withdraw credit from a client's wallet back into mine

POST

/reseller/v1/clients/{clientId}/debit

curl --request POST \
  --url https://api.nsin.cloud/reseller/v1/clients/1/debit \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --header 'Idempotency-Key: example' \
  --data '{ "amount_rials": 2000000, "description": "example" }'

• Production

The reverse of credit: moves Rials out of the client’s wallet and back into the calling reseller’s. Use it to reclaim funds you advanced — a client who cancels, a top-up sent to the wrong account, a trial you are winding down.

Mechanically it is the same transfer as credit with the two parties swapped, so it inherits the same guarantees: one debit and one credit in a single transaction or neither, both wallet rows locked FOR UPDATE in ascending id order so concurrent transfers in opposite directions cannot deadlock, and no ledger row at all when it is refused.

It cannot take a client’s balance below zero. There is no overdraft on this path in either direction. Asking for more than the client holds is 402 and nothing moves — so a client whose wallet is already negative from historical traffic billing cannot be drained further.

It does not check where the money came from. If a client funded his own wallet, this will take it. That is a deliberate limitation, not an oversight: NSIN does not track which Rial came from whom, and refusing on that basis would need a provenance ledger that does not exist. Treat the endpoint as the privileged operation it is.

It does not cancel anything. Withdrawing the money behind an active subscription does not end that subscription — the plan is already paid for and runs to its expiry. Cancel it with DELETE /services/{serviceId} if that is what you mean.

Both ledger rows carry ref_type: reseller_transfer, the same as credit. The direction is the sign on amount_rials.

Authorizations

• bearerAuth

Parameters

Path Parameters

clientId

required

integer

Must be a client of the caller, or the response is 403.

Header Parameters

Idempotency-Key

required

string

<= 191 characters

Required on every endpoint that moves money.

A client-generated unique string (a UUID is fine). Replaying the same key returns the original response byte for byte and does not repeat the operation. A replayed response is marked with the response header Idempotent-Replay: true.

This is not optional politeness. Without it, a network timeout on a credit or createService call leaves the caller unable to tell whether the money moved, and a naive retry double-charges.

A key never expires. There is no TTL and nothing prunes the table: the record is written inside the same transaction as the money movement and then kept forever, so a key is burned permanently. The same key sent a year later still replays the original answer instead of transferring again. Generate one key per operation — never one per process, per day, or per client.

Keys are scoped to the calling reseller, so two partners may pick the same UUID without colliding, and neither can probe the other’s key space.

Reuse with a different request is refused rather than replayed: same key, different method, path or body is 422. See the IdempotencyConflict response.

A failed operation records nothing, because the record and the money share one transaction and roll back together. A transfer that bounced for insufficient funds is therefore freely retryable with the same key once the wallet is funded.

The limit is 191 characters, not the 128 this document used to claim — it is the width of the database column, and it is the only validation performed on the value. The check runs on the trimmed string and counts bytes, so a non-ASCII key runs out before 191 characters; over-length is 400.

Request Body ^required

Media type application/json

object

amount_rials

required

Must be positive — the direction is the endpoint, not the sign. A negative amount is rejected rather than quietly treated as a credit, because a sign flip in a caller’s code must not be able to turn a withdrawal into a payment.

integer format: int64

>= 1

description

Shown on both ledger rows.

string

<= 191 characters

Responses

200

Withdrawn — or, if this Idempotency-Key was already used for this exact request, the first call’s response replayed unchanged.

Media type application/json

object

amount_rials

required

integer format: int64

reseller_balance_rials

required

My balance after the transfer.

integer format: int64

client_balance_rials

required

The client’s balance after the transfer.

integer format: int64

Example ^generated

{
  "amount_rials": 1,
  "reseller_balance_rials": 1,
  "client_balance_rials": 1
}

Headers

Idempotent-Replay

string

Allowed values: true

Sent as true only on a replay: the body is the stored answer from the first call with this Idempotency-Key, and the operation did NOT run a second time.

400

A malformed body, a non-positive amount, or a missing Idempotency-Key.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}

401

Missing, malformed, expired, or revoked credential.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}

402

The client does not hold that much. Nothing moved, no ledger row was written, and no idempotency key was recorded — retry with the same key for a smaller amount, or read GET /clients/{clientId}/wallet for the balance first.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}

403

Authenticated, but the target does not belong to you — or you are not a reseller at all.

This is the response the tenant-isolation matrix asserts on. Every route in this document is called with reseller B’s credential against reseller A’s ids, and must answer 403 or 404 with none of A’s data in the body.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}

404

No such resource.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}

422

This Idempotency-Key has already been used, for a different request.

A replay is only a replay if the method, path and body all match: they are hashed together into a fingerprint stored beside the recorded response, and a key that comes back with a different fingerprint is a caller bug — most often a key generated once per process instead of once per operation.

422 rather than 409 because the request is well-formed and authorized and no resource is in a conflicting state; it is the key that cannot be processed. Replaying the stored answer instead would tell the caller his SECOND transfer succeeded when it never ran, and he would report a client as funded who is not.

Nothing here is retryable until the key changes. The original operation stands, the new one never ran, and the fix is a fresh key.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}

429

The caller’s request budget is spent. 300 requests per minute by default (RESELLER_RATE_LIMIT), and the same limiter covers every one of the 83 operations in this document.

Counted per credential, not per reseller: each nsin_live_ token has its own budget and a dashboard session has another, so one runaway integration cannot lock its owner out of his own panel, and revoking that token is enough to stop it.

Branch on "code": "rate_limited" in the body. The message beside it is written for a human and may be reworded.

Retry-After is set, in seconds until the window resets — wait that long rather than retrying at once. The X-RateLimit-* headers are not on this response; they appear only on the responses the limiter let through, so a client that reads its remaining budget from the 429 alone will never see one.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}

Headers

Retry-After

integer

Seconds until the current window resets.

500

Something failed on our side. The body carries a human sentence and never an internal detail — a live sweep of this API once returned dial tcp 127.0.0.1:9000: connect: connection refused, which is our topology rather than an error message. That is now impossible.

Distinguish it from 503. A 503 means a dependency is down and the identical request will succeed later, so retry it. A 500 means the request hit a genuine fault: retrying it unchanged will fail the same way, and it should be reported with the X-Request-Id from the response header.

Declared on every operation because every operation can reach it. It was previously declared on exactly one, which left a generated client with no branch for the answer it is most likely to be surprised by.

Media type application/json

Every error body carries error, a human-readable sentence. Some also carry code, a stable machine-readable reason — branch on that, never on the sentence, which is prose and gets reworded.

object

error

required

string

code

Present only on the failures worth branching on, and deliberately not an exhaustive enum: treat a code you do not recognise as if it were absent and fall back to the status code.

The ones that exist today:

• credit_limit_reached — 402 from POST /services and POST /services/{serviceId}/change-plan. Your wallet, not the client’s: funding the purchase would take you past your overdraft (default 0, i.e. past zero). Top up; do not credit the client.
• domain_disabled — 409 from a configuration write on a disabled domain (records, rules, cache, SSL, members, settings). The domain is read-only until POST /domains/{domainId}/enable.
• unknown_rule_type — 404 from the rules paths when ruleType is not one of the documented values. You get this rather than an empty array, so a typo cannot read as a rule set that happens to be empty.
• rate_limited — 429 from any route. Keyed per credential, so one runaway integration cannot lock its owner out of the dashboard.
• session_check_failed — 503, and only on calls made with a dashboard session JWT. The session could not be verified, which is not the same as knowing it is revoked, so it is retryable and the session survives. Machine tokens never see this.

string

Example

{
  "error": "forbidden",
  "code": "credit_limit_reached"
}