Quickstart

Run your first QitOS minimal coding agent in under 2 minutes, then inspect it with qita.

This guide takes you from a fresh environment to a model-backed minimal coding agent and a live trace view in under 2 minutes.

Here is what the runtime looks like when you watch it work:

Need a provider first? Read Prerequisites. It explains how to get an OpenAI-compatible API key and validate your endpoint before running QitOS.

Prerequisites

• Python 3.10 or later
• pip
• An OpenAI-compatible API key

Steps

1. Install QitOS

   pip install "qitos[models]"
   
2. Set your provider config

   export OPENAI_API_KEY="sk-..."
   export OPENAI_BASE_URL="https://api.siliconflow.cn/v1/"
   export QITOS_MODEL="Qwen/Qwen3-8B"
   

   Any OpenAI-compatible endpoint works. Point OPENAI_BASE_URL at your provider if you’re not using OpenAI directly.

3. Run the minimal coding agent

   qit demo minimal
   

   This demo creates a small workspace with a buggy file, configures a real model, asks a minimal coding agent to fix the bug, runs verification, and writes a qita-ready trace. You should see output similar to:

   model_name: Qwen/Qwen3-8B
   workspace: .../playground/minimal_coding_agent
   target_file: buggy_module.py
   trace_run: .../runs/qitos_minimal_coding_<timestamp>
   final_result: Patch applied and verification passed.
   stop_reason: final
   next_step: qita board --logdir runs
   
4. Inspect the run with qita

   qita board --logdir runs
   

   Open http://127.0.0.1:8765 in your browser. The board shows every run in your log directory with step counts, stop reasons, and final results. Click view on any run to inspect its step-by-step trajectory (the temporal sequence of prompts, decisions, tool calls, and observations), or replay to play it back.

   The packaged demo and examples/quickstart/minimal_agent.py follow the same minimal coding-agent path. Start from the example file when you want to read or customize the code directly.

What’s next

You’ve run a minimal coding agent and inspected its trace. Here’s where to go from here:

Build your first agent

A walkthrough of the minimal coding-agent shape: state, model, tools, run, and qita

Tutorial track

Learn agent design step by step through ReAct, PlanAct, Claude Code-style, and security audit lessons

Observability

Everything qita can do: board, replay, and export

Benchmarks

Run GAIA, Tau-Bench, and CyBench on the QitOS kernel