The descriptions of the operations and fields are the contract’s own, in English.
/models/v1/familiesList model families
- Credential
- API key
- Permission
models.read
Lists all model families. Returns null when the catalog is empty.
curl "https://console.us.ccsio.ai/models/v1/families" \
-H "Authorization: Bearer $CCSIO_API_KEY"Parameters
| Name | In | Description |
|---|---|---|
X-Request-IDoptional | header | string Optional request ULID. A valid ULID (upper or lowercase) is honored; when the header is absent or invalid, the runtime mints the id. The response X-Request-ID header carries the canonical uppercase form. |
Responses
- 200
Success.
Schema
ListFamiliesResponseDisplayNamestringrequiredCapabilitiesarray of string or nullrequiredAllowed values:
"chat-completions""embeddings""responses""audio-speech""audio-transcriptions""image-generations""decisions"CreatedAtstring (date-time)requiredRFC3339, PostgreSQL timestamptz precision (microseconds); the offset is process-local — canonical UTC-Z is a deferred runtime gap (Amendment E).
IDstringrequiredpattern ^[A-Za-z0-9][A-Za-z0-9._:/-]{0,127}$
- 401
The request carries no credential, more than one, or one the platform does not accept, and is refused before the operation runs; `error.code` says why. Supply exactly one credential. Codes: invalid-request (no credential, or both Authorization:
Read more
Bearer and x-api-key), api-key-invalid (a value that is not a platform API key, or a key of a kind the platform does not accept), api-key-unknown, api-key-revoked, api-key-expired, bootstrap-disabled (the one-shot installation credential, once a key exists), token-malformed (a bearer token that is not a JWT), token-algorithm-unsupported, token-signature-invalid, token-claim-invalid, token-issuer-mismatch, token-audience-mismatch, token-expired, token-not-yet-valid, issuer-untrusted, issuer-residency-mismatch (the issuer belongs to another plane), issuer-resolution-unknown, jwks-unavailable, jwks-key-not-found, staff-token-inactive (a platform operator's token), staff-not-operator, staff-directory-unavailable, identity-not-provisioned (a verified identity with no account on this plane), organization-not-found (a verified identity or key whose organization this plane does not know), internal-error (a fault while the credential is checked is refused too, fail closed), context-cancelled
- 402
The organization has to pay first: its standing is payment_required.
Read more
The request boundary applies this payment gate to every route except an organization's billing routes, the caller's own context, organization list, onboarding, session report and personal keys, the public routes, and the routes of platform operators, who administer an organization that has to pay. The operations the gate covers declare this response. On /v1 the refusal is rendered in the OpenAI error format. Code: payment-required
- 403
No role of the caller holds `models.read` in the organization (policy-denied). Codes: policy-denied, organization-selection-required (a platform operator who selected no organization)
- 405
The path is claimed but the request method is not allowed. Go ServeMux returns its standard plain-text response and lists permitted methods in Allow.
- 500
Platform failure, rendered without internal cause details.
/models/v1/families/{id}Get a model family
- Credential
- API key
- Permission
models.read
Retrieves a model family by ID.
curl "https://console.us.ccsio.ai/models/v1/families/{id}" \
-H "Authorization: Bearer $CCSIO_API_KEY"Parameters
| Name | In | Description |
|---|---|---|
X-Request-IDoptional | header | string Optional request ULID. A valid ULID (upper or lowercase) is honored; when the header is absent or invalid, the runtime mints the id. The response X-Request-ID header carries the canonical uppercase form. |
idrequired | path | string Model family identifier (C0 ref). |
Responses
- 200
Success.
Schema
ModelFamilyDisplayNamestringrequiredCapabilitiesarray of string or nullrequiredAllowed values:
"chat-completions""embeddings""responses""audio-speech""audio-transcriptions""image-generations""decisions"CreatedAtstring (date-time)requiredRFC3339, PostgreSQL timestamptz precision (microseconds); the offset is process-local — canonical UTC-Z is a deferred runtime gap (Amendment E).
IDstringrequiredpattern ^[A-Za-z0-9][A-Za-z0-9._:/-]{0,127}$
- 400
Client refusal. This module answers a missing resource as a client refusal, not as 404: a family that does not exist is 400 family-not-found. Codes: ref-invalid (the family id is outside the C0 ref pattern), family-not-found
- 401
The request carries no credential, more than one, or one the platform does not accept, and is refused before the operation runs; `error.code` says why. Supply exactly one credential. Codes: invalid-request (no credential, or both Authorization:
Read more
Bearer and x-api-key), api-key-invalid (a value that is not a platform API key, or a key of a kind the platform does not accept), api-key-unknown, api-key-revoked, api-key-expired, bootstrap-disabled (the one-shot installation credential, once a key exists), token-malformed (a bearer token that is not a JWT), token-algorithm-unsupported, token-signature-invalid, token-claim-invalid, token-issuer-mismatch, token-audience-mismatch, token-expired, token-not-yet-valid, issuer-untrusted, issuer-residency-mismatch (the issuer belongs to another plane), issuer-resolution-unknown, jwks-unavailable, jwks-key-not-found, staff-token-inactive (a platform operator's token), staff-not-operator, staff-directory-unavailable, identity-not-provisioned (a verified identity with no account on this plane), organization-not-found (a verified identity or key whose organization this plane does not know), internal-error (a fault while the credential is checked is refused too, fail closed), context-cancelled
- 402
The organization has to pay first: its standing is payment_required.
Read more
The request boundary applies this payment gate to every route except an organization's billing routes, the caller's own context, organization list, onboarding, session report and personal keys, the public routes, and the routes of platform operators, who administer an organization that has to pay. The operations the gate covers declare this response. On /v1 the refusal is rendered in the OpenAI error format. Code: payment-required
- 403
No role of the caller holds `models.read` in the organization (policy-denied). Codes: policy-denied, organization-selection-required (a platform operator who selected no organization)
- 405
The path is claimed but the request method is not allowed. Go ServeMux returns its standard plain-text response and lists permitted methods in Allow.
- 500
Platform failure, rendered without internal cause details.
{
"ID": "glm",
"DisplayName": "GLM",
"Capabilities": [
"chat-completions",
"responses"
],
"CreatedAt": "2026-09-01T07:30:00Z"
}/models/v1/modelsList the model catalog
- Credential
- API key
- Permission
models.read
Lists every regional model with its family's display name and a summary of the caller's own organization's deployments of it. The organization is the verified principal's tenant, never a request field: deployments of any other organization are never listed.
Read more
The platform organization's deployments are shared with every organization that holds model access to the model, but they are not the caller's own and are never listed here, so an entry's deployments array does not say whether the model is served to the caller: GET /v1/models lists what the caller can call. For a human platform operator (a human whose user holds the active platform role ccsio_admin; models.read is part of their operator set) the tenant is the organization they selected with X-Organization-Id, any existing organization, so the console can read another organization's catalog view. A deployment summary carries what a catalog browser needs (lifecycle status, provider class, effective upstream protocol, upstream model, serving capabilities, residency and data region) and never the endpoint, the credential reference or the key identity. A model the organization has not deployed itself lists an empty deployments array. A model whose serving pool for the caller holds an active deployment served from a third-party provider connection carries the optional hosting member (kind third_party, the connection's display name and the processing location it declares), the disclosure the console shows before a person uses the model; it is omitted for every other model. Models are sorted by id and deployments by id; the response never renders null. A model an organization added to its own provider connection (`byo:<org_id>:<name>`) is listed only in the list of the organization that owns it: no other organization's list carries it.
curl "https://console.us.ccsio.ai/models/v1/models" \
-H "Authorization: Bearer $CCSIO_API_KEY"Parameters
| Name | In | Description |
|---|---|---|
X-Request-IDoptional | header | string Optional request ULID. A valid ULID (upper or lowercase) is honored; when the header is absent or invalid, the runtime mints the id. The response X-Request-ID header carries the canonical uppercase form. |
Responses
- 200
Success.
Schema
ListModelsResponsemodelsarray of ModelCatalogEntryrequiredidstringrequiredThe model ref a request names in its model field.
pattern ^[A-Za-z0-9][A-Za-z0-9._:/-]{0,127}$
family_idstringrequiredpattern ^[A-Za-z0-9][A-Za-z0-9._:/-]{0,127}$
family_namestringrequiredThe family's display name.
residencystringrequiredpattern ^[a-z][a-z0-9._-]{0,63}$
data_regionstringrequiredpattern ^[a-z][a-z0-9._-]{0,63}$
capabilitiesarray of stringrequiredWhat the regional model declares it serves. May be empty; a deployment's own capabilities say what it serves.
Allowed values:
"chat-completions""embeddings""responses""audio-speech""audio-transcriptions""image-generations""decisions"reasoninganyrequiredThe reasoning efforts a model takes as `reasoning_effort` (chat completions) and `reasoning.effort` (responses). `efforts` lists every value the model accepts, in the order a picker shows them, and is an empty array when the model takes none. `default` is what the model does when a request names no effort: one of `efforts`, or null when unknown (and always null when `efforts` is empty). It is advisory: the platform never adds an effort to a request that names none, and never filters or rewrites one a client sends.
created_atstring (date-time)requiredRFC3339, PostgreSQL timestamptz precision.
deploymentsarray of ModelDeploymentSummaryrequiredThe caller's organization's OWN deployments of this model, sorted by id; empty when it owns none. Deployments of another organization that serve the caller (the platform organization's, shared with every organization that holds model access to the model) are never listed here: whether the model can be called is answered by GET /v1/models.
hostingobjectoptionalWho processes a request to this model when it is not the platform itself: present only when the model's serving pool for the caller (the caller's own deployments when it has an active one, else the platform organization's in the caller's placement) holds an active deployment served from a third-party provider connection (provider class EXTERNAL_MANAGED with a stored key), omitted otherwise, so the entry of an ordinary model is exactly what it was before this member existed. It carries only the connection's public disclosure fields, never its base URL, the state of its key or any credential. When the pool holds several such deployments the entry names the connection of the deployment with the smallest id, the one selection prefers.
kindstringrequiredThe only kind today: the model is served by a third party's API, under the platform's own agreement with it.
Allowed values:
"third_party"provider_namestringrequiredThe connection's display name, as the operator who connected it typed it.
1 to 80 characters
processing_locationstringoptionalWhere the provider processes the data, as declared by the operator who connected it (free text); omitted when the connection declares none.
1 to 120 characters
- 400
Client refusal. Codes: invalid-request, ref-invalid.
- 401
The request carries no credential, more than one, or one the platform does not accept, and is refused before the operation runs; `error.code` says why. Supply exactly one credential. Codes: invalid-request (no credential, or both Authorization:
Read more
Bearer and x-api-key), api-key-invalid (a value that is not a platform API key, or a key of a kind the platform does not accept), api-key-unknown, api-key-revoked, api-key-expired, bootstrap-disabled (the one-shot installation credential, once a key exists), token-malformed (a bearer token that is not a JWT), token-algorithm-unsupported, token-signature-invalid, token-claim-invalid, token-issuer-mismatch, token-audience-mismatch, token-expired, token-not-yet-valid, issuer-untrusted, issuer-residency-mismatch (the issuer belongs to another plane), issuer-resolution-unknown, jwks-unavailable, jwks-key-not-found, staff-token-inactive (a platform operator's token), staff-not-operator, staff-directory-unavailable, identity-not-provisioned (a verified identity with no account on this plane), organization-not-found (a verified identity or key whose organization this plane does not know), internal-error (a fault while the credential is checked is refused too, fail closed), context-cancelled
- 402
The organization has to pay first: its standing is payment_required.
Read more
The request boundary applies this payment gate to every route except an organization's billing routes, the caller's own context, organization list, onboarding, session report and personal keys, the public routes, and the routes of platform operators, who administer an organization that has to pay. The operations the gate covers declare this response. On /v1 the refusal is rendered in the OpenAI error format. Code: payment-required
- 403
No role of the caller holds `models.read` in the organization (policy-denied). Codes: policy-denied, organization-selection-required (a platform operator who selected no organization)
- 405
The path is claimed but the request method is not allowed. Go ServeMux returns its standard plain-text response and lists permitted methods in Allow.
- 500
Platform failure, rendered without internal cause details.
{
"models": [
{
"id": "embed-1",
"family_id": "embed",
"family_name": "Embeddings",
"residency": "eu",
"data_region": "eu-central-1",
"capabilities": [
"embeddings"
],
"created_at": "2026-09-01T08:00:00Z",
"deployments": [],
"hosting": {
"kind": "third_party",
"provider_name": "Example AI",
"processing_location": "European Union"
},
"reasoning": {
"efforts": [],
"default": null
}
},
{
"id": "glm-5.3",
"family_id": "glm",
"family_name": "GLM",
"residency": "eu",
"data_region": "eu-central-1",
"capabilities": [
"chat-completions",
"responses"
],
"created_at": "2026-09-01T08:00:00Z",
"deployments": [
{
"id": "dep-glm-eu-1",
"status": "ACTIVE",
"provider_class": "CCSIO_INFERENCE",
"protocol": "openai-compat",
"upstream_model": "glm-5.3",
"capabilities": [
"chat-completions",
"responses"
],
"residency": "eu",
"data_region": "eu-central-1",
"created_at": "2026-09-01T09:00:00Z"
}
],
"reasoning": {
"efforts": [
"low",
"medium",
"high"
],
"default": "medium"
}
}
]
}/models/v1/models/{id}Get a regional model
- Credential
- API key
- Permission
models.read
Retrieves a regional model by ID. A model that another organization added to its own provider connection (`byo:<org_id>:<name>`) is not found, as if it did not exist.
curl "https://console.us.ccsio.ai/models/v1/models/{id}" \
-H "Authorization: Bearer $CCSIO_API_KEY"Parameters
| Name | In | Description |
|---|---|---|
X-Request-IDoptional | header | string Optional request ULID. A valid ULID (upper or lowercase) is honored; when the header is absent or invalid, the runtime mints the id. The response X-Request-ID header carries the canonical uppercase form. |
idrequired | path | string Regional model identifier (C0 ref). |
Responses
- 200
Success.
Schema
RegionalModelFamilystringrequiredpattern ^[A-Za-z0-9][A-Za-z0-9._:/-]{0,127}$
Residencystringrequiredpattern ^[a-z][a-z0-9._-]{0,63}$
DataRegionstringrequiredpattern ^[a-z][a-z0-9._-]{0,63}$
Capabilitiesarray of string or nullrequiredAllowed values:
"chat-completions""embeddings""responses""audio-speech""audio-transcriptions""image-generations""decisions"CreatedAtstring (date-time)requiredRFC3339, PostgreSQL timestamptz precision (microseconds); the offset is process-local — canonical UTC-Z is a deferred runtime gap (Amendment E).
IDstringrequiredpattern ^[A-Za-z0-9][A-Za-z0-9._:/-]{0,127}$
reasoninganyrequiredThe reasoning efforts a model takes as `reasoning_effort` (chat completions) and `reasoning.effort` (responses). `efforts` lists every value the model accepts, in the order a picker shows them, and is an empty array when the model takes none. `default` is what the model does when a request names no effort: one of `efforts`, or null when unknown (and always null when `efforts` is empty). It is advisory: the platform never adds an effort to a request that names none, and never filters or rewrites one a client sends.
- 400
Client refusal. This module answers a missing resource as a client refusal, not as 404: a model that does not exist, or that another organization added to its own provider connection, is 400 regional-model-not-found. Codes: ref-invalid (the model id is outside the C0 ref pattern), regional-model-not-found
- 401
The request carries no credential, more than one, or one the platform does not accept, and is refused before the operation runs; `error.code` says why. Supply exactly one credential. Codes: invalid-request (no credential, or both Authorization:
Read more
Bearer and x-api-key), api-key-invalid (a value that is not a platform API key, or a key of a kind the platform does not accept), api-key-unknown, api-key-revoked, api-key-expired, bootstrap-disabled (the one-shot installation credential, once a key exists), token-malformed (a bearer token that is not a JWT), token-algorithm-unsupported, token-signature-invalid, token-claim-invalid, token-issuer-mismatch, token-audience-mismatch, token-expired, token-not-yet-valid, issuer-untrusted, issuer-residency-mismatch (the issuer belongs to another plane), issuer-resolution-unknown, jwks-unavailable, jwks-key-not-found, staff-token-inactive (a platform operator's token), staff-not-operator, staff-directory-unavailable, identity-not-provisioned (a verified identity with no account on this plane), organization-not-found (a verified identity or key whose organization this plane does not know), internal-error (a fault while the credential is checked is refused too, fail closed), context-cancelled
- 402
The organization has to pay first: its standing is payment_required.
Read more
The request boundary applies this payment gate to every route except an organization's billing routes, the caller's own context, organization list, onboarding, session report and personal keys, the public routes, and the routes of platform operators, who administer an organization that has to pay. The operations the gate covers declare this response. On /v1 the refusal is rendered in the OpenAI error format. Code: payment-required
- 403
No role of the caller holds `models.read` in the organization (policy-denied). Codes: policy-denied, organization-selection-required (a platform operator who selected no organization)
- 405
The path is claimed but the request method is not allowed. Go ServeMux returns its standard plain-text response and lists permitted methods in Allow.
- 500
Platform failure, rendered without internal cause details.
{
"ID": "glm-5.3",
"Family": "glm",
"Residency": "eu",
"DataRegion": "eu-central-1",
"Capabilities": [
"chat-completions",
"responses"
],
"CreatedAt": "2026-09-01T08:00:00Z",
"reasoning": {
"efforts": [
"low",
"medium",
"high"
],
"default": "medium"
}
}Generated from the platform contracts at commit 5954fbc1.