Overview
The Anthropic harness provides direct integration with Anthropic’s Messages API. It supports Claude’s extended thinking feature for models like Claude Sonnet 4.5 and Claude 3.7 Sonnet.
Import
import { createGeneratorHarness, anthropicHarness } from "@llm-gateway/ai/harness/providers/anthropic";
Function Signature
function createGeneratorHarness(
apiKeyOrOptions?: string | AnthropicHarnessOptions
): GeneratorHarnessModule
Parameters
apiKeyOrOptions
string | AnthropicHarnessOptions
API key string or configuration objectAnthropic API key. Falls back to ANTHROPIC_API_KEY environment variable
Default model to use if not specified at invoke time
Returns
A harness module with invoke() and supportedModels() methods
What It Does
The Anthropic harness makes single LLM API calls and yields events for:
- reasoning: Streamed thinking content (for supported models)
- text: Streamed text content
- tool_call: Tool use from the model
- error: Any errors that occur
It does NOT:
- Execute tools (that’s the agent wrapper’s job)
- Handle permissions (that’s the agent wrapper’s job)
- Loop after tool calls (that’s the agent wrapper’s job)
Basic Example
import { createGeneratorHarness } from "@llm-gateway/ai/harness/providers/anthropic";
// Create with API key
const harness = createGeneratorHarness({
apiKey: process.env.ANTHROPIC_API_KEY,
model: "claude-sonnet-4-20250514",
});
// Make a single LLM call
for await (const event of harness.invoke({
model: "claude-sonnet-4-20250514",
messages: [{ role: "user", content: "Hello!" }],
})) {
if (event.type === "text") {
console.log(event.content);
}
}
Using the Singleton
import { anthropicHarness } from "@llm-gateway/ai/harness/providers/anthropic";
// Use pre-configured singleton (reads ANTHROPIC_API_KEY from env)
for await (const event of anthropicHarness.invoke({
model: "claude-opus-4-20250514",
messages: [{ role: "user", content: "Explain quantum computing" }],
})) {
if (event.type === "text") {
console.log(event.content);
}
}
Extended Thinking
Claude Sonnet 4.5 and Claude 3.7 Sonnet support extended thinking:
for await (const event of harness.invoke({
model: "claude-sonnet-4-5-20250929",
messages: [{ role: "user", content: "Solve this complex problem..." }],
})) {
if (event.type === "reasoning") {
console.log("[Thinking]", event.content);
}
if (event.type === "text") {
console.log("[Answer]", event.content);
}
}
- Higher max tokens (16000 vs 4096)
- Thinking budget of 8000 tokens
import { z } from "zod";
const tools = [
{
name: "search",
description: "Search the web for information",
schema: z.object({
query: z.string(),
}),
},
];
for await (const event of harness.invoke({
model: "claude-sonnet-4-20250514",
messages: [{ role: "user", content: "What are the latest AI developments?" }],
tools,
})) {
if (event.type === "tool_call") {
console.log(`Tool: ${event.name}`);
console.log(`Input:`, event.input);
}
}
System Messages
System messages are automatically extracted and passed to Anthropic’s system parameter:
for await (const event of harness.invoke({
model: "claude-sonnet-4-20250514",
messages: [
{ role: "system", content: "You are a helpful coding assistant" },
{ role: "user", content: "Write a Python function to sort a list" },
],
})) {
if (event.type === "text") {
console.log(event.content);
}
}
Multimodal Support
Supports images and PDF documents:
for await (const event of harness.invoke({
model: "claude-sonnet-4-20250514",
messages: [
{
role: "user",
content: [
{ type: "text", text: "Analyze this document" },
{
type: "document",
mediaType: "application/pdf",
data: base64PdfData,
},
],
},
],
})) {
if (event.type === "text") {
console.log(event.content);
}
}
// This message structure:
[
{ role: "assistant", tool_calls: [{ id: "1", name: "search", arguments: {...} }] },
{ role: "tool", tool_call_id: "1", content: "Result 1" },
{ role: "tool", tool_call_id: "2", content: "Result 2" },
]
// Becomes:
[
{ role: "assistant", content: [{ type: "tool_use", ... }] },
{ role: "user", content: [
{ type: "tool_result", tool_use_id: "1", content: "Result 1" },
{ type: "tool_result", tool_use_id: "2", content: "Result 2" },
]},
]
List Available Models
const models = await harness.supportedModels();
console.log("Available models:", models);
// Dynamically fetches from Anthropic API or falls back to known models
Wrapping with Agent Harness
import { createAgentHarness } from "@llm-gateway/ai/harness/agent";
import { createGeneratorHarness } from "@llm-gateway/ai/harness/providers/anthropic";
// Wrap Anthropic provider with agent capabilities
const agent = createAgentHarness({
harness: createGeneratorHarness(),
maxIterations: 10,
});
// Now supports tool execution and looping
for await (const event of agent.invoke({
model: "claude-sonnet-4-20250514",
messages: [{ role: "user", content: "Research and summarize AI news" }],
tools: [searchTool, readTool],
})) {
console.log(event);
}
Error Handling
for await (const event of harness.invoke({
model: "claude-sonnet-4-20250514",
messages: [{ role: "user", content: "Hello!" }],
})) {
if (event.type === "error") {
console.error("Error:", event.error.message);
}
}
Thinking-Enabled Models
These models automatically get extended thinking enabled:
claude-sonnet-4-5-20250929claude-3-7-sonnet-20250219
Other models use standard inference without thinking blocks.
Architecture Notes
- Uses Anthropic SDK for streaming
- Tracks tool calls by content block index
- Handles thinking deltas separately from text deltas
- Automatically converts between message formats
- Single iteration only - compose with agent harness for loops
