AI Provider

An AI Provider records one remote LLM endpoint. It does not keep a URL, a header, a TLS setting, or a credential. A related Nautobot External Integration owns all of those. This app does no inference. It only makes a catalog of what is available.

Each AI Provider owns zero or more AI Models.

The class is AIProvider, not Provider. Nautobot core already has circuits.Provider. One unqualified name for two things confuses a reader and an import.

Fields

Field	Type	Required	Description
name	string	yes	The unique name of the provider.
description	string	no	Free text.
external_integration	foreign key to extras.ExternalIntegration	yes	Gives the remote URL, the headers, the TLS settings, the timeout, and the credentials. Protected: Nautobot refuses to delete an External Integration that a provider uses.
provider_type	choice	yes	The API dialect of this endpoint: openai, anthropic, openai_compatible, or ollama. The default is openai. A consuming app reads this field to find out how to address the provider.
openai_compatible	boolean	yes	The endpoint gives the OpenAI API shape, which includes GET /v1/models. The default is True. The Discover AI Models job runs only against a provider that has this flag set.
enabled	boolean	yes	Whether this provider is in service. The default is True. Discovery skips a disabled provider. An app that reads this registry must also skip one.
num_predict	integer	no	The default maximum number of tokens to generate. -1 means unlimited. An AI Model can override it.
temperature	decimal	no	The default sampling temperature, from 0 to 2. An AI Model can override it.

Provider type and OpenAI-compatible

The two fields answer different questions. You need both.

• openai_compatible answers can this app discover models here. The Discover AI Models job reads it, because GET /v1/models is the only discovery endpoint this app knows.
• provider_type answers how does a client address this endpoint. A consuming app reads it before the app sends anything.

Ollama shows the difference. Ollama gives an OpenAI-compatibility layer, so openai_compatible is a true statement about Ollama. But that layer does not return tool calls in the tool_calls field. A model that gets a request for a tool writes the JSON call into the message content, where nothing can act on it. The native Ollama API returns tool calls correctly. An app that reads only the boolean addresses Ollama through the compatibility layer and loses tool calling silently. Set provider_type to ollama and openai_compatible to True. This records both facts.

Two of the four types are an address, not a service:

Type	Needs a remote URL
openai	no
anthropic	no
openai_compatible	yes
ollama	yes

A self-hosted vLLM or llama.cpp endpoint has no well-known address.

WARNING: A client with no URL for one of these two types goes to a default endpoint. That endpoint is another company's API, and the client sends this provider's credential to it. Give an openai_compatible or an ollama provider an External Integration with a remote URL. The app refuses to save one without a URL.

Providers that existed before this field

The migration that added provider_type fills it in from openai_compatible. It answers for one case only.

Row before the migration	provider_type after
OpenAI-compatible, with a remote URL	openai_compatible
OpenAI-compatible, no remote URL	empty
Not OpenAI-compatible	empty

The boolean records only that an endpoint is OpenAI-shaped. It says nothing about what a row that is not OpenAI-shaped speaks instead, so the migration leaves that row empty.

The openai_compatible dialect needs a remote URL, so the migration does not give it to a row that has none. Such a row is left empty too. Otherwise the next save would fail over the External Integration, a field that the operator did not touch.

The app refuses to save an empty value. Thus an operator must answer the next time the operator edits the record. The form shows an empty option for such a row, so no save can write a dialect that nobody chose.

How to take a provider out of service

Clear the Enabled checkbox. Each of these is a provider-level event, and this field records it: a contract that ends, a machine that goes down for maintenance, a key that changes, or spend that must stop now.

CAUTION: The Discover AI Models job creates a model with Enabled set. A provider that you take out of service one model at a time comes back on the next discovery run. Do not disable each model. Clear the flag on the provider.

CAUTION: AIModel.provider cascades. If you delete a provider, you also delete each model record on it, together with the cost data. Do not delete a provider that you want to keep. Clear the flag on it.

To find out if one model is on offer, read AIModel.is_available. It asks one question in place of two.

Credentials

Attach a Secrets Group to the External Integration. Keep the API key as a secret with the access type HTTP(S) and the secret type token. The app reads the key at the point of use. It never keeps the value.