> ## Documentation Index
> Fetch the complete documentation index at: https://docs.deepmako.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Mako-32B Conductor

> The flagship model powering autonomous tool orchestration and complex reasoning.

## Overview

| Property           | Value                                          |
| ------------------ | ---------------------------------------------- |
| **Parameters**     | 32 billion                                     |
| **Architecture**   | Decoder-only transformer (Qwen3-32B fine-tune) |
| **Context Window** | 32,768 tokens                                  |
| **Tool Calling**   | Native structured tool calling                 |
| **Request format** | `model: "conductor"`                           |

The Conductor is Mako's flagship model. It excels at multi-step tool orchestration — resolving ENS names, fetching balances, calculating USD values, chaining web searches, and synthesizing results into natural responses.

This is the model that powers [deepmako.com](https://deepmako.com) and the production API.

## How it works

When a request arrives, the gateway routes it to the Conductor model and enters the **agentic tool loop**:

1. The model receives the conversation history, system prompt, and tool definitions.
2. If the model responds with tool calls, the gateway executes them and feeds the results back.
3. This loop runs for up to **8 rounds**, allowing Mako to chain complex multi-step research autonomously.
4. The final text response is returned in OpenAI format.

All tool orchestration is handled server-side — your application receives a single final answer.

## Inference parameters

| Parameter   | Default |
| ----------- | ------- |
| Temperature | 0.4     |
| Top-p       | 0.9     |

Conservative sampling ensures reliable tool calling and consistent output quality.

## Cold starts

The first request after a period of inactivity may take 30–90 seconds while the model loads. The gateway handles this automatically with polling. If you're building a frontend, the `503` status code indicates a cold start — display a loading state and retry.

## When to use

* Complex queries requiring multiple tool calls
* Tasks that need reasoning over fetched data
* Production accuracy-critical workflows
* Any query where quality matters more than speed