Providers

client.providers stores the model keys your agents run on. You save a key once as a provider, then point any number of agents at it with providerId and a model. To choose a provider type and model, read Providers and models.

const provider = await client.providers.create({
  name: "OpenRouter",
  providerType: "openrouter",
  baseUrl: null,
  apiKey: process.env.OPENROUTER_API_KEY!,
});

const { models } = await client.providers.listModels({ providerId: provider.id });
console.log(models.some(({ id }) => id === "openai/gpt-6-luna"));

Every method takes one input object and accepts an optional abortSignal. Blazing Agents never returns a key after you save it; responses show only its last four characters in keyFragment.

Available operations

Methods

create()

Saves a provider key.

Signature: create(input: CreateProviderBody & ResourceRequestOptions): Promise<ProviderResponse>

const provider = await client.providers.create({
  name: "OpenRouter",
  providerType: "openrouter",
  baseUrl: null,
  apiKey: process.env.OPENROUTER_API_KEY!,
});

Use "custom" with a baseUrl for any other OpenAI-compatible endpoint. The type, key, and base URL cannot change later, so create a new provider to rotate a key. Returns ProviderResponse. Errors: validation_failed, provider_name_conflict, provider_limit_reached.

list()

Lists your providers.

Signature: list(input?: ResourceRequestOptions): Promise<ProvidersResponse>

const { providers } = await client.providers.list();

Returns { providers: ProviderListItem[] }. List items have id, name, providerType, createdAt, and updatedAt; call get() for baseUrl and keyFragment.

get()

Reads one provider.

Signature: get(input: { providerId: string } & ResourceRequestOptions): Promise<ProviderResponse>

const provider = await client.providers.get({ providerId });

Returns ProviderResponse. Errors: validation_failed, not_found.

listModels()

Lists the model IDs the provider offers right now. It runs no model and costs nothing.

Signature: listModels(input: { providerId: string } & ResourceRequestOptions): Promise<ProviderModelsResponse>

const { models } = await client.providers.listModels({ providerId });

Returns { models: Array<{ id: string }> }, sorted by ID. Pass one of these IDs as an agent's model; agent create and update check the model against this same list.

For "vercel_ai_gateway", the list shows what the gateway offers, not what your key can use: credits, team policy, or routing can still make a turn fail. "custom" providers raise model_discovery_unsupported, and you type the model ID yourself. Errors: not_found, model_discovery_unsupported, model_validation_unavailable.

getThinkingLevels()

Lists the reasoning levels a model accepts for an agent's thinkingLevel.

Signature: getThinkingLevels(input: { providerId: string; model: string } & ResourceRequestOptions): Promise<ThinkingLevelsResponse>

const { known, levels } = await client.providers.getThinkingLevels({
  providerId,
  model: "openai/gpt-6-luna",
});
if (known && levels.includes("high")) {
  await client.agents.update({ agentId, thinkingLevel: "high" });
}

Returns { known: boolean; levels: string[] }. When known is true, levels lists every accepted value, and an empty list means only the provider default (null) works. When known is false, Blazing Agents does not know the model, so any value is accepted but the provider may reject it during a turn. It works for any model ID, including custom providers.

update()

Renames a provider. The name is the only field you can change.

Signature: update(input: UpdateProviderBody & { providerId: string } & ResourceRequestOptions): Promise<ProviderResponse>

const provider = await client.providers.update({ providerId, name: "OpenRouter production" });

Returns ProviderResponse. Errors: validation_failed, not_found, provider_name_conflict.

delete()

Deletes a provider and its key.

Signature: delete(input: DeleteProviderOptions & { providerId: string } & ResourceRequestOptions): Promise<void>

await client.providers.delete({ providerId, confirmVersionInvalidation: true });

Deletion fails with provider_in_use while any agent's current configuration uses the provider; move those agents first. It fails with provider_historical_use when only old versions, pinned sessions, or tasks refer to it; details lists them. Pass confirmVersionInvalidation: true to delete anyway. History stays readable, but running or restoring those versions then fails with provider_not_found.

Response types

ProviderResponse

Errors

Failures throw BlazingAgentsError. The provider codes:

Next

• Providers and models
• Agents reference