Start free →

Error codes

Model providers and the trial key

Each code of this family: what it means, its most frequent cause, what to do and who can do it.

All error codes

provider_key_rejected

What it means
The model provider refused the API key you entered, so it was not saved.
Most frequent cause
A key revoked at the provider, copied incompletely, or from another account or project.
What to do
The provider rejected this API key, so it was not saved. Copy it again from the provider's console — it may be revoked or incomplete.
Who can do it
An owner or admin of the workspace · Open the screen

provider_permission_denied

What it means
The model provider accepted the workspace's API key but refused this call (HTTP 403).
Most frequent cause
The key's account or project has no access to the model, the model is not offered in its region, or the key lacks a permission the call needs (a restricted or read-only key).
What to do
The provider accepted the key but refused this call: its account or project has no access to this model, in this region, or lacks a permission. Grant the access in the provider's console, or pick another model — replacing the key will not help.
Who can do it
An owner or admin of the workspace

provider_key_unusable

What it means
The text entered is not an API key the provider accepts for model calls, so it was not saved.
Most frequent cause
An admin key, or the sign-in token of a subscription (a chat app, a coding tool), instead of a standard API key — or a key copied with a space, a line break or another character inside it.
What to do
This is not an API key that can make model calls, so it was not saved — an admin key, a sign-in token, or a copy with a space, a line break or a stray character inside. Create a standard API key in the provider's console and paste it alone.
Who can do it
An owner or admin of the workspace · Open the screen

provider_key_storage_unavailable

What it means
Azoth could not encrypt the key, so it did not store it: this deployment cannot store provider keys right now.
Most frequent cause
A configuration problem on the deployment’s side — the encryption setting for stored keys is missing or invalid — not a problem with the key. Outside SaaS, the deployment’s operator fixes it.
What to do
The key was not saved: this deployment cannot store provider keys right now. Nothing is wrong with your key — contact support so it can be fixed, then save it again.
Who can do it
Azoth, or the operator of this deployment

provider_not_connected

What it means
The call goes to a provider the workspace has no API key for.
Most frequent cause
The provider was never connected, or its key was removed.
What to do
No API key for this provider is connected to the workspace. Add it in Settings › Providers and models.
Who can do it
An owner or admin of the workspace · Open the screen

provider_model_not_found

What it means
The provider does not offer this model, or your key cannot use it.
Most frequent cause
A mistyped or retired model name, or a model your account with the provider has no access to.
What to do
The provider does not offer this model, or your key cannot use it. Check the model name, or pick another model.
Who can do it
The developer of the calling code

azure_config_required

What it means
A call to Azure OpenAI names neither the resource nor the deployment it is for, so it cannot be sent.
Most frequent cause
The azureEndpoint or azureDeployment field is missing from the request, or a rule routes the call to Azure OpenAI while the caller did not send them.
What to do
A call to Azure OpenAI must name its resource and deployment: send azureEndpoint and azureDeployment with it.
Who can do it
The developer of the calling code

azure_endpoint_not_allowed

What it means
The azureEndpoint of a call to Azure OpenAI is not an Azure OpenAI host, so the workspace's key is not sent to it.
Most frequent cause
The endpoint names another host, a port, a user name or plain http — or a proxy in front of Azure.
What to do
Azure OpenAI is reached at its own hosts only: set azureEndpoint to https://<resource>.openai.azure.com (or your Azure AI services host).
Who can do it
The developer of the calling code

invalid_azure_deployment

What it means
The azureDeployment of a call to Azure OpenAI is not a deployment name Azoth can put in the request.
Most frequent cause
The name holds a /, a ?, .. or a space, or is longer than 64 characters.
What to do
An Azure deployment name holds letters, digits, dots, dashes and underscores only, 64 at most: send the deployment's name as Azure shows it.
Who can do it
The developer of the calling code

endpoint_url_not_allowed

What it means
An address Azoth was asked to call — an Ollama base URL, an audit-export destination, an S3 endpoint — is refused before any request: it is not a public address.
Most frequent cause
The address is on a private network, the loopback or a link-local range, resolves to one, or uses another scheme than http(s); the answer's reason names which.
What to do
Azoth does not call this address: give a public https address — no private network, loopback or link-local host, no user name in it. The answer's reason says what was refused.
Who can do it
The developer of the calling code

ollama_endpoint_missing

What it means
This call was blocked: this workspace has no Ollama address.
Most frequent cause
The gateway calls Ollama only at the address the workspace saved in Settings › Providers and models, and none is saved.
What to do
Ask an owner or admin to save the address of your Ollama server in Settings › Providers and models, then send the call again.
Who can do it
An owner or admin of the workspace · Open the screen

ollama_endpoint_not_declared

What it means
This call was blocked: it named an Ollama address this workspace did not save.
Most frequent cause
The gateway calls Ollama only at the address saved in Settings › Providers and models; the call’s ollamaBaseUrl is another one.
What to do
Leave ollamaBaseUrl out of the call, or send exactly the address saved. To use another server, an owner or admin saves its address there.
Who can do it
The developer of the calling code · Open the screen

provider_rate_limited

What it means
The provider refused the call because of a rate limit or a credit limit on your account with it.
Most frequent cause
Too many calls in a short time, or no credit left with the provider.
What to do
The provider refused the call: rate limit reached or no credit left on your account with them. Check your plan and billing with the provider, then retry.
Who can do it
An owner or admin of the workspace

provider_payment_required

What it means
The model provider refused the call for a billing reason (HTTP 402, or a refusal its error names as one).
Most frequent cause
The provider account of the workspace's API key has no credit left, its payment method failed or expired, it reached a spending limit, or its plan was cancelled.
What to do
The provider refused the call for a billing reason: its account behind this workspace's key has no credit left or no valid payment method, or has reached a spending limit. Add credit, fix the billing or raise the limit in the provider's console, then send the call again.
Who can do it
An owner or admin of the workspace

provider_request_too_large

What it means
The model provider refused the call because the request is larger than it accepts for the model (HTTP 413).
Most frequent cause
Too many or too long messages, or a large attachment, in one call.
What to do
The provider refused the call: the request is larger than it accepts for this model. Send fewer or shorter messages or attachments, then send the call again.
Who can do it
The developer of the calling code

provider_request_rejected

What it means
The model provider refused the call without serving it (HTTP 400, 422 or another 4xx), for a reason a retry does not fix.
Most frequent cause
Most often a provider account with no credit left — some providers refuse every call with a 400 then — else a parameter the model does not accept: an option, a message format, or messages longer than the model's context.
What to do
The provider refused the call, and retrying will not help until the cause is fixed. Check first the credit and billing of its account behind this workspace's key, in the provider's console — some providers refuse every call this way when the account has no credit left — then the call's parameters: the model's options, the format and the length of the messages.
Who can do it
An owner or admin of the workspace

provider_not_supported

What it means
This call was blocked: a rule sends it to a provider this endpoint cannot reach.
Most frequent cause
A rule of your workspace reroutes this call to ‹provider›, which this endpoint does not support.
What to do
Use the /api/v1/chat endpoint, or ask the rule’s owner to change where it reroutes.
Who can do it
An owner or admin of the workspace · Open the screen

provider_failure

What it means
The provider did not answer.
Most frequent cause
The model provider returned an error. Azoth billed nothing.
What to do
Try again, or choose another model.
Who can do it
The developer of the calling code

onboarding_no_provider

What it means
This first call was not sent: no model provider is connected to the workspace.
Most frequent cause
The first call goes to your own provider, with your own key — Azoth sends no call without one.
What to do
Connect a provider in Settings › Providers and models (or in the form on this page), then send it again.
Who can do it
An owner or admin of the workspace · Open the screen

trial_exhausted

What it means
The Azoth trial key answers no more calls for this workspace.
Most frequent cause
Its trial calls are used up: the trial key is there for a first try, not for regular use.
What to do
Connect your own provider key in Settings › Providers and models: your calls then go through Azoth exactly as before.
Who can do it
An owner or admin of the workspace · Open the screen

trial_unavailable

What it means
The Azoth trial key cannot answer this call.
Most frequent cause
The trial key serves workspaces on Free or in trial only, within a daily limit shared by all of them, and only while Azoth’s own AI is running.
What to do
Connect your own provider key in Settings › Providers and models, or try again later.
Who can do it
An owner or admin of the workspace · Open the screen

trial_limits_exceeded

What it means
This call is outside what the Azoth trial key answers.
Most frequent cause
The trial key answers short, non-streamed calls to its one model only.
What to do
Send a shorter call without streaming, to the trial’s model, or connect your own provider key in Settings › Providers and models.
Who can do it
The developer of the calling code · Open the screen

trial_not_eligible

What it means
The Azoth trial key is offered only to workspaces on Free or in their trial.
Most frequent cause
The workspace is on a paid plan.
What to do
The Azoth trial key is for workspaces on Free or in trial.
Who can do it
An owner or admin of the workspace · Open the screen

trial_already_used

What it means
The Azoth trial key is offered once per person.
Most frequent cause
You turned it on in another workspace already.
What to do
You already used an Azoth trial key in another workspace. Connect your own provider key in Settings › Providers and models.
Who can do it
An owner or admin of the workspace · Open the screen

egress_blocked

What it means
This call was blocked: this deployment cannot reach ‹provider›.
Most frequent cause
This Azoth runs without Internet access and may only reach the hosts its operator allowed; ‹provider› is not one of them.
What to do
Use a model the deployment hosts internally, or ask the operator whether ‹provider› should be allowed. This setting belongs to the deployment, not to a rule.
Who can do it
The developer of the calling code

secnumcloud_blocked

What it means
This call was blocked: the provider is not SecNumCloud-certified.
Most frequent cause
This deployment only allows SecNumCloud-certified providers, and ‹provider› is not one of them.
What to do
Use a certified provider offered below. This setting belongs to the deployment, not to a rule.
Who can do it
The developer of the calling code