Agent SDK

Agent loops — multi-turn tool use with state — run on the standard chat-completions endpoint today. A first-party agent SDK with managed loops and tool execution is on the roadmap.

TypeScript

// Minimal agent loop with plain chat completions — no framework needed.
// Tools are defined per the tool-calling guide; loop until no tool_calls.
while (true) {
  const resp = await client.chat.completions.create({
    model: "anthropic/claude-sonnet-5",
    messages,
    tools,
  });
  const msg = resp.choices[0].message;
  messages.push(msg);
  if (!msg.tool_calls?.length) break;
  for (const call of msg.tool_calls) {
    const result = await runTool(call.function.name, call.function.arguments);
    messages.push({ role: "tool", tool_call_id: call.id, content: result });
  }
}

The loop above is the whole pattern: call, execute tools, feed results back, repeat until the model stops calling tools. Frameworks like Mastra or Pydantic AI wrap the same mechanics — see the framework guides.