Anvia

@anvia/anthropic API reference ​

Import every public symbol from @anvia/anthropic. The package has no public subpath exports.

Direct Anthropic client ​

ts
type AnthropicClientOptions =
  | { apiKey: string; baseUrl?: string; client?: never }
  | { client: Anthropic; apiKey?: never; baseUrl?: never }

type AnthropicCompletionModelOptions = {
  modelId: AnthropicCompletionModelId
  contextLimits?: ModelContextLimits
  controls?: Controls
}

class AnthropicClient implements ModelListingClient {
  constructor(options: AnthropicClientOptions)
  completionModel(options: AnthropicCompletionModelOptions):
    AnthropicCompletionModelHandle
  listModels(options?: { abortSignal?: AbortSignal }): Promise<ModelList>
}

All constructors and factories require an options object. Managed credentials and an injected SDK client are mutually exclusive. The completion model ID is explicit; there is no factory default.

AnthropicCompletionModelHandle is a StreamingCompletionModel<unknown, Controls>. It normalizes Messages API responses, stream events, usage, tool calls, documents, images, and reasoning into Anvia completion contracts. completionModel() accepts an optional controls object on both clients. When omitted, controls resolve from the model ID through the exported AnthropicControlsFor type, which tiers a typed reasoning-effort control drawn from ANTHROPIC_REASONING_EFFORTS (low, medium, high, xhigh, max): claude-fable-5, claude-mythos-5, claude-opus-4-7, claude-opus-4-8, claude-opus-5, and claude-sonnet-5 accept all five efforts; claude-mythos-preview, claude-opus-4-6, and claude-sonnet-4-6 accept up to max; claude-opus-4-5 accepts up to high; unknown IDs expose no controls. The selected effort is forwarded as output_config.effort on the Messages request.

Vertex AI client ​

ts
type AnthropicVertexClientOptions =
  | (Omit<ClientOptions, 'maxRetries'> & { client?: never })
  | { client: AnthropicVertex /* all managed options are excluded */ }

type AnthropicVertexCompletionModelOptions = {
  modelId: AnthropicCompletionModelId
  contextLimits?: ModelContextLimits
  controls?: Controls
}

class AnthropicVertexClient {
  constructor(options: AnthropicVertexClientOptions)
  completionModel(options: AnthropicVertexCompletionModelOptions):
    AnthropicVertexCompletionModelHandle
}

ClientOptions comes from @anthropic-ai/vertex-sdk. Vertex and injected-client construction are mutually exclusive, the model ID is required, and this client intentionally has no listModels() method.

Model IDs and namespace ​

AnthropicCompletionModelId is ModelId<KnownAnthropicCompletionModelId>, preserving autocomplete for known Claude IDs while accepting other string IDs. The exact known-ID union is exported and evolves with provider support.

The anthropic namespace re-exports the same clients and public option, handle, and model-ID types. It has no state of its own.

Built for Anvia.