@anvia/anthropic

Anthropic provider adapter for Anvia.

Use this package when you want Anvia agents, extractors, or pipelines to run on Claude models through Anthropic's SDK, or through an Anthropic-compatible API endpoint.

Installation

pnpm add @anvia/anthropic @anvia/core

In this monorepo, the package is available through the workspace:

pnpm --filter @anvia/anthropic build

Usage

import { Agent } from "@anvia/core";
import { AnthropicClient } from "@anvia/anthropic";

const client = new AnthropicClient({
  apiKey,
});

const model = client.completionModel({ modelId: "claude-opus-5" });

const agent = new Agent({
  id: "assistant",
  model: model,
  instructions: "Answer clearly and concisely.",
});

const result = await agent.generate({ prompt: "Summarize Anvia in one sentence." });
if (result.type === "response") console.log(result.output);

Anthropic-Compatible APIs

For APIs that expose an Anthropic-compatible surface, pass a custom baseUrl:

import { AnthropicClient } from "@anvia/anthropic";

const client = new AnthropicClient({
  apiKey,
  baseUrl,
});

const model = client.completionModel({ modelId: "provider/model-name" });

Vertex AI

Use Anthropic's official Vertex SDK through AnthropicVertexClient:

import { AnthropicVertexClient } from "@anvia/anthropic";

const client = new AnthropicVertexClient({
  projectId: "my-gcp-project",
  region: "global",
});

const model = client.completionModel({ modelId: "claude-sonnet-5" });

The client follows the standard Google authentication flow. It reads ANTHROPIC_VERTEX_PROJECT_ID and CLOUD_ML_REGION when project and region are omitted, and supports Application Default Credentials such as GOOGLE_APPLICATION_CREDENTIALS:

export ANTHROPIC_VERTEX_PROJECT_ID="my-gcp-project"
export CLOUD_ML_REGION="global"
export GOOGLE_APPLICATION_CREDENTIALS="/absolute/path/service-account.json"

For custom authentication, pass the official SDK's googleAuth, authClient, or accessToken options:

pnpm add google-auth-library

import { GoogleAuth } from "google-auth-library";

const client = new AnthropicVertexClient({
  projectId: "my-gcp-project",
  region: "global",
  googleAuth: new GoogleAuth({
    credentials: serviceAccountJson,
    scopes: "https://www.googleapis.com/auth/cloud-platform",
  }),
});

You can also pass a preconfigured authClient, including impersonated credentials. Validate externally supplied credential configurations before using them, and never commit credential JSON or service-account keys.

Vertex AI does not expose Anthropic's Models API, so AnthropicVertexClient intentionally does not provide listModels().

Exports

• AnthropicClient
• AnthropicVertexClient
• structural Anthropic and Vertex completion handle types
• AnthropicCompletionModelId
• anthropic

Development

pnpm --filter @anvia/anthropic typecheck
pnpm --filter @anvia/anthropic test
pnpm --filter @anvia/anthropic build

Package-local typecheck and build scripts build @anvia/core first so core subpath types are available in a fresh worktree.