Thinking & Reasoning¶

Many modern models (like Claude 3.7 Sonnet, OpenAI o1, or Gemini Pro) support thinking/reasoning capabilities where they can show their internal thought process before responding.

Requesting Reasoning¶

You can enable thinking modes through the provider options passed to the stream or complete APIs.

import { llm, stream, Message } from "@codeworksh/aikit";

const model = await llm("anthropic", "claude-3-7-sonnet-20250219");

const context: Message.Context = {
    messages: [
        {
            role: "user",
            time: { created: Date.now() },
            parts: [{ type: "text", text: "Solve this complex logic puzzle..." }],
        },
    ],
};

// Start a stream with thinking enabled
const s = stream(model, context, {
    thinkingEnabled: true,
    thinkingBudgetTokens: 4096,
});

Streaming Thinking Content¶

When streaming, thinking content is delivered separately from standard text output. You can catch this via the reasoning.delta (or equivalent based on Vercel AI SDK integration) event type.

for await (const event of s) {
    if (event.type === "reasoning.delta") {
        process.stdout.write(event.delta); // Stream the model's internal thoughts
    } else if (event.type === "text.delta") {
        process.stdout.write(event.delta); // Stream the final text response
    }
}

The final message object will include reasoning content directly in its parts if the model emitted any.