Hosted providers
List the providers this deployment serves hosted inference on.
Stored providers and those defined in the configuration's hosted_providers: block alike, each marked with its source.
Enter your API key here (sent as Otari-Key header).
0100Successful Response
A page of hosted providers, and how many there are in total.
Validation Error
GET /api/v1/hosted-providers HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
{
"count": 1,
"data": [
{
"api_base": "text",
"api_key_last4": "text",
"client_args": {
"ANY_ADDITIONAL_PROPERTY": "anything"
},
"created_at": "2026-01-01T00:00:00.000Z",
"decryptable": true,
"enabled": true,
"has_api_key": true,
"id": "123e4567-e89b-12d3-a456-426614174000",
"name": "text",
"provider": "text",
"restricted_to_organization_id": "123e4567-e89b-12d3-a456-426614174000",
"serves_unlisted_models": false,
"shadows_config": false,
"source": "database",
"supported_countries": [
"text"
],
"updated_at": "2026-01-01T00:00:00.000Z",
"zero_data_retention": false
}
]
}Configure a provider with the key this deployment will serve it on.
Everything the provider lists on that key is offered at once; a provider that will not say yields a provider with no models, not an error. A provider the configuration defines may be created too, and the stored one then serves instead.
Enter your API key here (sent as Otari-Key header).
Configure a provider this deployment will serve hosted inference on.
provider is the any-llm implementation. name is what requests
address it by (name:model), defaulting to provider; a name of its
own lets two endpoints of one implementation coexist. api_key may be
left out for a backend that needs none, or for an endpoint of its own
(api_base), which is called with a placeholder key.
client_args is whatever the provider's SDK client needs beyond the key
(Bedrock's region and IAM pair), the same shape an organization provider
key takes. Values under credential-shaped names come back masked.
restricted_to_organization_id keeps the provider to one organization;
every other one is served as if it were not configured.
serves_unlisted_models serves a model the roster does not list, as a
config.yml instance does, and switches on every model it offers
whether or not something prices it. supported_countries keeps the
provider to requests from those ISO country codes, checked as well as
provider_supported_countries; null serves every country.
zero_data_retention declares that the upstream keeps no prompt or
completion data; the catalog shows it and nothing enforces it.
truefalsefalseSuccessful Response
One hosted provider, known by its name, its implementation and the tail of its key.
No shape in this section carries key material: a stored credential is
known by its last four characters and nothing else, client_args goes
out with credential-shaped values masked, and api_base with the
password in its userinfo and its query values masked.
name is what requests address and every route takes; provider is
the any-llm implementation. source says where the provider is defined.
A config provider comes from the hosted_providers: block, has no
row and so no id or timestamps, and is changed by editing the
configuration rather than through this API; its models are administered as
a stored provider's are. shadows_config marks a stored provider that
wins over a configuration entry of the same name, which serves again once
the row is deleted. has_api_key is false for one stored without a key,
and decryptable false for one whose key or extras no configured
OTARI_SECRET_KEY reads, which serves nothing until re-entered.
serves_unlisted_models is true for one that serves a model its roster
does not list. supported_countries is null for one that serves every
country. zero_data_retention is the operator's declaration that the
upstream keeps no prompt or completion data.
truetruefalsefalsedatabasePossible values: falseValidation Error
POST /api/v1/hosted-providers HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 55
{
"api_key": "<your OpenAI API key>",
"provider": "openai"
}{
"api_base": "text",
"api_key_last4": "text",
"client_args": {
"ANY_ADDITIONAL_PROPERTY": "anything"
},
"created_at": "2026-01-01T00:00:00.000Z",
"decryptable": true,
"enabled": true,
"has_api_key": true,
"id": "123e4567-e89b-12d3-a456-426614174000",
"name": "text",
"provider": "text",
"restricted_to_organization_id": "123e4567-e89b-12d3-a456-426614174000",
"serves_unlisted_models": false,
"shadows_config": false,
"source": "database",
"supported_countries": [
"text"
],
"updated_at": "2026-01-01T00:00:00.000Z",
"zero_data_retention": false
}Apply the reviewed prices, then refresh every provider's models, then the cleanup if asked.
The price step comes first, so the model step prices new models at the prices it applied, and a model held for want of a price is served once it has one. Without a digest no price changes. A digest that no longer names the pending prices answers 409 and changes nothing. cleanup removes the prices of models no provider serves, with their history.
Enter your API key here (sent as Otari-Key header).
What to apply, as the preview showed it.
Also remove the prices of models no hosted provider serves, with their history.
falseThe digest of the previewed prices, to apply them; 409 when the pending prices differ. Without it, no price changes.
Successful Response
What a catalog update did.
Validation Error
POST /api/v1/hosted-providers/catalog-update HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 33
{
"cleanup": false,
"digest": "text"
}{
"catalog": {
"kept": [
{
"count": 1,
"models": [
"text"
],
"reason": "text"
}
],
"orphaned_rosters": [
"text"
],
"providers": [
{
"added": [
"text"
],
"credential_unreadable": false,
"discovery_unsupported": false,
"error": "text",
"provider": "text",
"repriced": [
"text"
],
"source": "database"
}
],
"removed": [
"text"
],
"removed_count": 1,
"removed_override_rows": 1,
"removed_price_rows": 1
},
"prices_applied": true
}What updating the catalog would do: the models.dev price changes, then what each provider newly lists.
A POST, because it holds the fetched prices as pending for the apply to name by their digest; nothing else is written. It dials every enabled provider. A price fetch that fails is reported in price_error and the model lists are still previewed. With cleanup, it also previews which prices of models no provider serves would be removed.
Enter your API key here (sent as Otari-Key header).
What to preview: the update alone, or with the cleanup.
Also preview removing the prices of models no hosted provider serves.
falseSuccessful Response
What a catalog update would do: its price changes, then what each provider newly lists.
prices is None when models.dev changed nothing, or when it could not
be fetched, which price_error then says. catalog carries the cleanup
the preview was asked for, and no removals otherwise.
Validation Error
POST /api/v1/hosted-providers/catalog-update/preview HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 17
{
"cleanup": false
}{
"catalog": {
"kept": [
{
"count": 1,
"models": [
"text"
],
"reason": "text"
}
],
"orphaned_rosters": [
"text"
],
"providers": [
{
"added": [
"text"
],
"credential_unreadable": false,
"discovery_unsupported": false,
"error": "text",
"provider": "text",
"repriced": [
"text"
],
"source": "database"
}
],
"removed": [
"text"
],
"removed_count": 1,
"removed_override_rows": 1,
"removed_price_rows": 1
},
"price_error": "text",
"prices": {
"added_count": 1,
"changed_count": 1,
"changes": [
{
"change": "added",
"model_key": "text"
}
],
"changes_truncated": true,
"digest": "text",
"fetched_at": "2026-01-01T00:00:00.000Z",
"needs_review": true,
"removed_count": 1,
"review_reason": "text"
}
}What moving the deployment providers the retired dashboard page stored onto hosted providers would do.
The gateway moves them at startup, so this lists the ones it could not move, each with the reason. Writes nothing and dials nothing.
Enter your API key here (sent as Otari-Key header).
Successful Response
The deployment providers the retired dashboard page stored, and how each moves onto a hosted provider.
GET /api/v1/hosted-providers/migration HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
Successful Response
{
"rows": [
{
"api_base": "text",
"detail": "text",
"instance": "text",
"models": 1,
"notes": [
"text"
],
"outcome": "migrate",
"provider": "text",
"source_removed": false,
"updated_at": "2026-01-01T00:00:00.000Z"
}
]
}Move the deployment providers the retired dashboard page stored onto hosted providers of the same name.
Each is created as a hosted provider that serves models not on its roster, offering what it lists as a create does, and the deployment provider is removed. One that cannot move is reported and left in place. rows limits the move to the rows of the preview the caller reviewed, at the versions it was shown them.
Enter your API key here (sent as Otari-Key header).
Which deployment providers to move: the ones the caller reviewed, as reviewed.
rows names the rows of the preview the caller saw, each with the
updated_at it was shown at. A row the preview did not show, or one
changed since, is left where it is, so nothing moves that nobody reviewed.
Omitted, every row that can move does.
Successful Response
The deployment providers the retired dashboard page stored, and how each moves onto a hosted provider.
Validation Error
POST /api/v1/hosted-providers/migration HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 71
{
"rows": [
{
"instance": "openai-eu",
"updated_at": "2026-10-19T09:00:00Z"
}
]
}{
"rows": [
{
"api_base": "text",
"detail": "text",
"instance": "text",
"models": 1,
"notes": [
"text"
],
"outcome": "migrate",
"provider": "text",
"source_removed": false,
"updated_at": "2026-01-01T00:00:00.000Z"
}
]
}Remove a deployment provider the move could not take, which serves nothing.
Enter your API key here (sent as Otari-Key header).
Successful Response
No content
Validation Error
DELETE /api/v1/hosted-providers/migration/{instance} HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
No content
Re-encrypt every stored key and set of client arguments with the primary OTARI_SECRET_KEY.
Run after rotating the key, alongside POST /api/v1/search-tools/reencrypt. A provider no configured key reads is left as it is, to be re-entered.
Enter your API key here (sent as Otari-Key header).
Successful Response
What re-encrypting the stored hosted provider credentials did.
unreadable counts providers no configured key reads, left as they were
to be re-entered; skipped those edited while the rotation ran, already
written under the primary key.
POST /api/v1/hosted-providers/reencrypt HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
Successful Response
{
"reencrypted": 1,
"skipped": 1,
"unreadable": 1
}Ask a provider what it lists on the given credentials, storing nothing.
Backs the form's "Test connection". With name naming a stored provider, a field left out is read from it, so an edit can be tested without typing the key again; pointing the stored key at another origin still needs it. Failure comes back in the body.
Enter your API key here (sent as Otari-Key header).
Credentials to try against a provider without storing them.
The fields the create body takes. name naming an existing stored
provider lets a field left out fall back to what it stores, so an edit form
can test without re-entering the key; pointing the stored key at another
origin still needs the key typed again, as an update does.
Successful Response
What a connection test found: whether the provider answered, and how many models it lists.
discovery_unsupported is set for a provider with no model listing,
which a test cannot reach that way although it may still serve.
false0Validation Error
POST /api/v1/hosted-providers/test HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 55
{
"api_key": "<your OpenAI API key>",
"provider": "openai"
}{
"discovery_unsupported": false,
"error": "text",
"model_count": 0,
"ok": true
}Rotate the key, repoint the base, or turn a provider off.
A provider defined in configuration answers 409: it is changed there.
Enter your API key here (sent as Otari-Key header).
Rotate the key, repoint the base, or turn a provider off.
provider and name are immutable: changing either is a delete plus a
create, because the key belongs to the provider it was issued by and
requests address the name. Omitting api_key leaves the stored
credential untouched, so a toggle never re-sends a secret the caller does
not have; an explicit null removes it, where the provider can be served
without one. client_args omitted is left alone; an explicit
null clears it, and an entry echoed back as the mask keeps the stored value.
restricted_to_organization_id reads the same way: omitted keeps the
restriction, null lifts it, and so does supported_countries, where an
empty list closes the provider to every country.
Successful Response
One hosted provider, known by its name, its implementation and the tail of its key.
No shape in this section carries key material: a stored credential is
known by its last four characters and nothing else, client_args goes
out with credential-shaped values masked, and api_base with the
password in its userinfo and its query values masked.
name is what requests address and every route takes; provider is
the any-llm implementation. source says where the provider is defined.
A config provider comes from the hosted_providers: block, has no
row and so no id or timestamps, and is changed by editing the
configuration rather than through this API; its models are administered as
a stored provider's are. shadows_config marks a stored provider that
wins over a configuration entry of the same name, which serves again once
the row is deleted. has_api_key is false for one stored without a key,
and decryptable false for one whose key or extras no configured
OTARI_SECRET_KEY reads, which serves nothing until re-entered.
serves_unlisted_models is true for one that serves a model its roster
does not list. supported_countries is null for one that serves every
country. zero_data_retention is the operator's declaration that the
upstream keeps no prompt or completion data.
truetruefalsefalsedatabasePossible values: falseValidation Error
PATCH /api/v1/hosted-providers/{provider} HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 17
{
"enabled": false
}{
"api_base": "text",
"api_key_last4": "text",
"client_args": {
"ANY_ADDITIONAL_PROPERTY": "anything"
},
"created_at": "2026-01-01T00:00:00.000Z",
"decryptable": true,
"enabled": true,
"has_api_key": true,
"id": "123e4567-e89b-12d3-a456-426614174000",
"name": "text",
"provider": "text",
"restricted_to_organization_id": "123e4567-e89b-12d3-a456-426614174000",
"serves_unlisted_models": false,
"shadows_config": false,
"source": "database",
"supported_countries": [
"text"
],
"updated_at": "2026-01-01T00:00:00.000Z",
"zero_data_retention": false
}Remove a provider and its roster, leaving the runtime with no hosted path for it.
A provider defined in configuration answers 409. A stored provider that overrides a configured one keeps the roster, which the configured one serves again.
Enter your API key here (sent as Otari-Key header).
Successful Response
A human-readable acknowledgment for an operation with nothing to return.
What happened.
Validation Error
DELETE /api/v1/hosted-providers/{provider} HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
{
"message": "text"
}Ask the provider what it serves on its credential.
Dials the upstream on every call rather than caching: the caller is the admin form's model picker, opened rarely and entitled to a current answer. Failure comes back in the body, so the picker can fall back to a text box.
Enter your API key here (sent as Otari-Key header).
Successful Response
What the provider says it serves on the stored credential.
Failure is a field rather than a status: an unreachable upstream, or a provider with no model listing, is an answer about the provider rather than about this request, and the form still has to render a plain text box when the list cannot be fetched.
falseValidation Error
GET /api/v1/hosted-providers/{provider}/available-models HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
{
"discovery_unsupported": false,
"error": "text",
"models": [
"text"
],
"provider": "text"
}List the models offered on one provider, with the price each currently serves at.
Enter your API key here (sent as Otari-Key header).
0500Only models of this kind; unknown for those no source names a kind for
Successful Response
A page of one hosted provider's offered models, and how many there are in total.
count is the total of the kind asked for, and type_counts how many
of each kind the provider offers, unknown for a model of no known kind.
Validation Error
GET /api/v1/hosted-providers/{provider}/models HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
{
"count": 1,
"data": [
{
"cache_read_price_per_million": 1,
"cache_write_price_per_million": 1,
"created_at": "2026-01-01T00:00:00.000Z",
"disabled_reason": "unpriced",
"enabled": true,
"id": "123e4567-e89b-12d3-a456-426614174000",
"input_price_per_million": 1,
"model": "text",
"output_price_per_million": 1,
"price_source": "organization",
"provider": "text",
"type": "chat",
"unit": "text",
"updated_at": "2026-01-01T00:00:00.000Z"
}
],
"type_counts": {
"ANY_ADDITIONAL_PROPERTY": 1
}
}Offer a model on a provider, optionally at a custom price.
Enter your API key here (sent as Otari-Key header).
Offer a model on a hosted provider, optionally at a custom price.
Omitting the rates means the deployment serves the model at whatever the
price ladder already answers: a stored deployment price, else the community
default, which the offer stores as the deployment's own rate. unit is
what the rates are per, as on POST /pricing: a request or image rate is
still stored per million.
Successful Response
One offered model, with the price the deployment currently serves it at.
price_source says which rung of the deployment's ladder answered:
deployment for a rate an operator set, defaults for the
community-maintained rate, whether this surface stored it or the fallback
supplies it, and None when nothing prices the model yet. unit says what
the rates are per, so a per-request rate is not read as a per-token one.
enabled is the serving switch; a model offered without a discoverable
price starts off. disabled_reason says why a model is off: unpriced
while it waits for a rate, operator when an operator switched it off,
unavailable when its provider refused to serve it. type is the
kind of model, from models.dev, and None when it does not know.
Validation Error
POST /api/v1/hosted-providers/{provider}/models HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 78
{
"input_price_per_million": 2.5,
"model": "gpt-4o",
"output_price_per_million": 10
}{
"cache_read_price_per_million": 1,
"cache_write_price_per_million": 1,
"created_at": "2026-01-01T00:00:00.000Z",
"disabled_reason": "unpriced",
"enabled": true,
"id": "123e4567-e89b-12d3-a456-426614174000",
"input_price_per_million": 1,
"model": "text",
"output_price_per_million": 1,
"price_source": "organization",
"provider": "text",
"type": "chat",
"unit": "text",
"updated_at": "2026-01-01T00:00:00.000Z"
}Ask the provider again and offer whatever is newly listed.
Nothing already offered is removed, even when the provider no longer lists it. New models follow the offer rule, seeded with the community default price and off when nothing prices them. A seeded price nobody changed moves with the default. An offered model whose stored price has since been deleted is seeded again, or switched off where the default does not price it, even one an operator switched on: a refresh never switches a model on.
Enter your API key here (sent as Otari-Key header).
Successful Response
What a refresh did: the models newly offered, the seeded rates it moved, and the list's new size.
Failure is a field rather than a status, for the reason
HostedAvailableModelsPublic gives: the list is still standing, and the
panel renders the reason beside it.
falseValidation Error
POST /api/v1/hosted-providers/{provider}/models/refresh HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
{
"added": [
"text"
],
"count": 1,
"discovery_unsupported": false,
"error": "text",
"repriced": [
"text"
]
}Reprice one offered model, toggle whether it is served, or both.
Enter your API key here (sent as Otari-Key header).
Reprice an offered model, toggle whether it is served, or both.
A price is the input/output pair or nothing: sending the pair writes a new
pricing version, and the cache rates then follow POST /pricing's own
semantics (a field the caller omits inherits the most recent stored value,
an explicit null clears it). An omitted unit carries the stored one
forward. enabled travels alone freely, so a toggle never re-sends rates
it did not change.
Successful Response
One offered model, with the price the deployment currently serves it at.
price_source says which rung of the deployment's ladder answered:
deployment for a rate an operator set, defaults for the
community-maintained rate, whether this surface stored it or the fallback
supplies it, and None when nothing prices the model yet. unit says what
the rates are per, so a per-request rate is not read as a per-token one.
enabled is the serving switch; a model offered without a discoverable
price starts off. disabled_reason says why a model is off: unpriced
while it waits for a rate, operator when an operator switched it off,
unavailable when its provider refused to serve it. type is the
kind of model, from models.dev, and None when it does not know.
Validation Error
PATCH /api/v1/hosted-providers/{provider}/models/{model_id} HTTP/1.1
Otari-Key: YOUR_API_KEY
Content-Type: application/json
Accept: */*
Content-Length: 17
{
"enabled": false
}{
"cache_read_price_per_million": 1,
"cache_write_price_per_million": 1,
"created_at": "2026-01-01T00:00:00.000Z",
"disabled_reason": "unpriced",
"enabled": true,
"id": "123e4567-e89b-12d3-a456-426614174000",
"input_price_per_million": 1,
"model": "text",
"output_price_per_million": 1,
"price_source": "organization",
"provider": "text",
"type": "chat",
"unit": "text",
"updated_at": "2026-01-01T00:00:00.000Z"
}Stop offering a model on a provider. Its pricing history stays.
Enter your API key here (sent as Otari-Key header).
Successful Response
A human-readable acknowledgment for an operation with nothing to return.
What happened.
Validation Error
DELETE /api/v1/hosted-providers/{provider}/models/{model_id} HTTP/1.1
Otari-Key: YOUR_API_KEY
Accept: */*
{
"message": "text"
}Last updated
Was this helpful?
