@tetherai/anthropic

Standalone Anthropic provider for TetherAI - Everything you need in one package!

This package provides a complete, streaming-first solution for the Anthropic Messages API.
No external dependencies required - includes all types, utilities, and middleware built-in.
Think of it as Express for AI providers with everything included.

What's Included

• Anthropic Provider: Streaming chat completions with full API support
• Enhanced Chat Options: Temperature, maxTokens, topP, topK, stop sequences, system prompts
• Non-Streaming Chat: Complete response handling for simple requests
• Model Management: List models, validate model IDs, get token limits
• Retry Middleware: Automatic retries with exponential backoff
• Fallback Middleware: Multi-provider failover support
• Error Handling: Rich error classes with HTTP status codes
• Edge Runtime: Works everywhere from Node.js to Cloudflare Workers
• SSE Utilities: Built-in Server-Sent Events parsing
• TypeScript: 100% typed with zero any types

Quick Start

Installation

npm install @tetherai/anthropic
# or
pnpm add @tetherai/anthropic
# or
yarn add @tetherai/anthropic

That's it! No additional packages needed - everything is included.

⚠️ Important: Anthropic API requires maxTokens parameter. If you don't provide it, the API will return a 400 error. Always include maxTokens in your requests.

Basic Usage

Set your API key:

export ANTHROPIC_API_KEY=sk-ant-...

Streaming Chat Example

import { anthropic } from "@tetherai/anthropic";

const provider = anthropic({ 
  apiKey: process.env.ANTHROPIC_API_KEY!,
  timeout: 30000,        // 30 second timeout
  maxRetries: 2          // Built-in retry configuration
});

for await (const chunk of provider.streamChat({
  model: "claude-3-5-sonnet-20240620",
  messages: [{ role: "user", content: "Hello!" }],
  temperature: 0.7,      // Enhanced chat options
  maxTokens: 1000,
  systemPrompt: "You are a helpful assistant."
})) {
  if (chunk.done) break;
  process.stdout.write(chunk.delta);
}

Non-Streaming Chat Example

const response = await provider.chat({
  model: "claude-3-5-sonnet-20240620",
  messages: [{ role: "user", content: "Hello!" }],
  temperature: 0.5,
  maxTokens: 500,
  responseFormat: "json_object"  // Get structured responses
});

console.log(response.content);
console.log(`Used ${response.usage.totalTokens} tokens`);

Model Management Example

// Get available models
const models = await provider.getModels();
console.log("Available models:", models);

// Validate model ID
const isValid = provider.validateModel("claude-3-5-sonnet-20240620");
console.log("Model valid:", isValid);

// Get token limits
const maxTokens = provider.getMaxTokens("claude-3-5-sonnet-20240620");
console.log("Max tokens:", maxTokens);

Next.js Edge Runtime Example

// app/api/chat/route.ts
import { NextRequest } from "next/server";
import { anthropic, withRetry } from "@tetherai/anthropic";

export const runtime = "edge";

const provider = withRetry(
  anthropic({ 
    apiKey: process.env.ANTHROPIC_API_KEY!,
    timeout: 30000,
    apiVersion: "2023-06-01", // API version control
  }), 
  { retries: 2 }
);

export async function POST(req: NextRequest) {
  const body = await req.json();
  const stream = provider.streamChat({
    model: "claude-3-5-sonnet-20240620",
    messages: body.messages,
    temperature: body.temperature || 0.7,
    maxTokens: body.maxTokens || 1000,
    systemPrompt: body.systemPrompt,
    stop: body.stopSequences,
    responseFormat: body.responseFormat
  });

  return new Response(new ReadableStream({
    async start(controller) {
      const encoder = new TextEncoder();
      for await (const chunk of stream) {
        controller.enqueue(encoder.encode(`data: ${JSON.stringify(chunk)}\n\n`));
      }
      controller.close();
    },
  }), {
    headers: { "Content-Type": "text/event-stream" },
  });
}

Enhanced Chat Options

Advanced Chat Configuration

const response = await provider.chat({
  model: "claude-3-5-sonnet-20240620",
  messages: [{ role: "user", content: "Write a story" }],
  
  // Core parameters
  temperature: 0.8,           // 0-1, controls randomness
  maxTokens: 1000,            // Maximum response length
  topP: 0.9,                  // 0-1, nucleus sampling
  topK: 40,                   // 0-100, top-k sampling (Claude specific)
  
  // Stop sequences
  stop: ["\n\n", "END"],      // Stop generation at these sequences
  
  // System behavior
  systemPrompt: "You are a creative storyteller", // Alternative to system messages
  
  // Response format
  responseFormat: "json_object", // Get structured JSON responses
  
  // Safety and moderation
  safeMode: true,              // Enable content filtering
  
  // Metadata
  user: "user123",             // User identifier for moderation
  metadata: {                  // Custom metadata
    sessionId: "abc123",
    source: "web"
  }
});

Parameter Mapping

| TS Interface Field | Anthropic API Field | |---------------------|----------------------------| | maxTokens | max_tokens | | topP | top_p | | topK | top_k | | responseFormat | response_format.type | | user | metadata.user | | metadata | metadata (merged) |

Note: user is mapped to metadata.user because Anthropic does not support a top‑level user field. Optional anthropic-beta header can be passed via beta field in options.

Middleware Compatibility

| Feature | Support | |----------------|---------| | withRetry | ✅ | | withFallback | ✅ |

Wrap the provider with withRetry(...) or withFallback([...]) to add resilience without changing your code.

Streaming with Enhanced Options

for await (const chunk of provider.streamChat({
  model: "claude-3-5-sonnet-20240620",
  messages: [{ role: "user", content: "Explain quantum physics" }],
  temperature: 0.3,            // More focused responses
  maxTokens: 2000,             // Longer explanation
  topP: 0.95,                 // High quality sampling
  topK: 50,                   // Claude-specific top-k sampling
  stop: ["\n\n", "In conclusion"], // Natural stopping points
  systemPrompt: "You are a physics professor explaining complex concepts simply"
})) {
  if (chunk.done) {
    console.log(`\nFinished: ${chunk.finishReason}`);
    console.log(`Usage: ${chunk.usage?.totalTokens} tokens`);
    break;
  }
  process.stdout.write(chunk.delta);
}

Configuration Options

Anthropic Provider Options

interface AnthropicOptions {
  apiKey: string;                    // Required: Your Anthropic API key
  baseURL?: string;                  // Custom API endpoint (default: https://api.anthropic.com/v1)
  apiVersion?: string;               // API version (default: 2023-06-01)
  maxRetries?: number;               // Maximum retry attempts
  timeout?: number;                  // Request timeout in ms (default: 30000)
  fetch?: Function;                  // Custom fetch implementation
}

Advanced Configuration

import { anthropic } from "@tetherai/anthropic";

const provider = anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY!,
  baseURL: "https://api.anthropic.com/v1",  // Custom endpoint
  apiVersion: "2023-06-01",                 // API version control
  timeout: 60000,                           // 60 second timeout
  maxRetries: 3,                            // 3 retry attempts
  fetch: customFetch                         // Custom fetch for proxies, etc.
});

Middleware System

Retry Middleware

Automatically retries failed requests with exponential backoff:

import { anthropic, withRetry } from "@tetherai/anthropic";

const provider = withRetry(
  anthropic({ apiKey: process.env.ANTHROPIC_API_KEY! }),
  {
    retries: 3,        // Number of retry attempts
    baseMs: 300,       // Base delay in milliseconds
    factor: 2,         // Exponential backoff factor
    jitter: true       // Add randomness to prevent thundering herd
  }
);

Smart Error Detection: Only retries on transient errors (429 rate limits, 5xx server errors)

Fallback Middleware

Chain multiple providers for automatic failover:

import { anthropic, withFallback, withRetry } from "@tetherai/anthropic";
import { openAI } from "@tetherai/openai";

const provider = withFallback([
  withRetry(anthropic({ apiKey: process.env.ANTHROPIC_API_KEY! }), { retries: 2 }),
  withRetry(openAI({ apiKey: process.env.OPENAI_API_KEY! }), { retries: 2 })
], {
  onFallback: (error, providerIndex) => {
    console.log(`Provider ${providerIndex} failed, trying next...`);
  }
});

Error Handling

Rich error classes with detailed information:

import { AnthropicError } from "@tetherai/anthropic";

try {
  await provider.streamChat(request);
} catch (error) {
  if (error instanceof AnthropicError) {
    console.log(`Anthropic error ${error.status}: ${error.message}`);
    
    // Handle specific error types
    switch (error.status) {
      case 401: // Invalid API key
        console.log("Please check your API key");
        break;
      case 429: // Rate limited
        console.log("Rate limited - will retry automatically");
        break;
      case 500: // Server error
        console.log("Anthropic server error - will retry automatically");
        break;
    }
  }
}

Edge Runtime Compatibility

Works seamlessly in modern edge environments:

• Next.js Edge Runtime
• Vercel Edge Functions
• Cloudflare Workers
• Deno Deploy
• Node.js (all versions)

Performance Features

• Streaming-First: Real-time token streaming with AsyncIterable
• Memory Efficient: No buffering of entire responses
• Automatic Retries: Built-in resilience for production use
• Edge Optimized: Uses native fetch and ReadableStream
• Enhanced Options: Full control over response generation
• Model Management: Built-in model validation and token limits
• Claude Optimized: Built specifically for Anthropic's API patterns
• Enterprise Ready: Custom fetch, comprehensive error handling

API Reference

Core Functions

• anthropic(options) → Creates Anthropic provider instance
• provider.streamChat(request) → AsyncIterable of chat chunks
• provider.chat(request) → Promise of complete chat response
• provider.getModels() → List available models
• provider.validateModel(modelId) → Check if model is supported
• provider.getMaxTokens(modelId) → Get token limit for model
• withRetry(provider, options) → Wraps provider with retry logic
• withFallback(providers, options) → Creates multi-provider failover

Types

• AnthropicOptions → Configuration interface
• AnthropicError → Error class with HTTP status
• ChatRequest → Enhanced chat completion request
• ChatStreamChunk → Streaming response chunk with metadata
• ChatResponse → Complete chat response with usage info
• Provider → Common provider interface
• ModelInfo → Model capabilities and pricing

Examples

See examples for ready-to-run demos:

• Next.js chat app – Full Edge runtime UI example
• Node.js server – Minimal backend SSE server

Why This Package?

• Zero Dependencies: Everything included, no external packages needed
• Production Ready: Built-in retry, fallback, and error handling
• Highly Configurable: Timeouts, custom endpoints, API version control
• Edge Compatible: Works everywhere from Node.js to Cloudflare Workers
• Streaming First: Real-time token streaming with AsyncIterable
• Enhanced Features: Full chat options, model management, non-streaming support
• Claude Optimized: Built specifically for Anthropic's API patterns
• Enterprise Ready: Custom fetch, comprehensive error handling

License

MIT