Providers

client.providers stores the model provider keys your agents run on. You send a key once; Blazing Agents keeps it and never returns it. You can then list the provider's models and check which reasoning levels a model supports, without running the model.

Examples assume client = BlazingAgents(). Every method also accepts extra_headers and timeout. On AsyncBlazingAgents, await the same method names.

import os

provider = client.providers.create(
    name="OpenRouter",
    provider_type="openrouter",
    api_key=os.environ["OPENROUTER_API_KEY"],
)
models = client.providers.list_models(provider.id).models
print([model.id for model in models][:5])

Available operations

Methods

create()

Stores a provider key under a name.

provider = client.providers.create(
    name="Internal gateway",
    provider_type="custom",
    api_key=os.environ["GATEWAY_API_KEY"],
    base_url="https://llm.example.com/v1",
)

Signature: create(*, name: str, provider_type: ProviderType, api_key: str, base_url=...) -> Provider

A missing base_url for "custom" or a non-None base_url for "vercel_ai_gateway" raises ValueError before any request. Never log api_key.

Returns Provider. Raises APIStatusError with validation_failed, provider_name_conflict, or provider_limit_reached.

list()

Lists your providers.

providers = client.providers.list().providers

Signature: list() -> Providers

Returns Providers, whose providers field is a list of ProviderListItem with id, name, provider_type, created_at, and updated_at. Use get() for the base URL and key fragment.

get()

Gets one provider.

provider = client.providers.get(provider.id)
print(provider.key_fragment)

Signature: get(provider_id: str) -> Provider

Returns Provider. Raises validation_failed or not_found.

list_models()

Lists the model IDs the provider offers. This never runs a model and costs nothing.

ids = [model.id for model in client.providers.list_models(provider.id).models]

Signature: list_models(provider_id: str) -> ProviderModels

Returns ProviderModels, whose models field lists ProviderModel items with an id, sorted and without duplicates. Use an id as the agent's model.

A listed model is not a guarantee that your key can use it: credits, account policy, or routing can still reject a request. "custom" providers raise model_discovery_unsupported; type their model IDs yourself. A temporarily unavailable catalog raises model_validation_unavailable.

get_thinking_levels()

Lists the reasoning levels one model supports, for an agent's thinking_level.

levels = client.providers.get_thinking_levels(provider.id, model="openai/gpt-6-luna")
if levels.known and "high" in levels.levels:
    client.agents.update("ag_0123456789abcdef", thinking_level="high")

Signature: get_thinking_levels(provider_id: str, *, model: str) -> ThinkingLevels

Returns ThinkingLevels with known: bool and levels: list[str]. When known is True, only the listed levels are accepted, and an empty list means the model supports only the provider default (thinking_level=None). When known is False, Blazing Agents has no data for the model and accepts any string, which the provider can still reject at run time.

update()

Renames a provider.

provider = client.providers.update(provider.id, name="OpenRouter production")

Signature: update(provider_id: str, *, name=...) -> Provider

Only the name can change. To change the type, key, or base URL, create a new provider and move your agents to it. Calling update() without name raises ValueError before any request. Returns Provider. Raises validation_failed, not_found, or provider_name_conflict.

delete()

Deletes a provider and its stored key.

client.providers.delete(provider.id, confirm_version_invalidation=True)

Signature: delete(provider_id: str, *, confirm_version_invalidation: bool = False) -> None

After a confirmed delete, history is kept, but running or restoring anything that needs the provider raises provider_not_found.

Response models

Provider

Next

• Providers and models
• Agents
• Run your first agent