Toprak1yu/agent-tool-use-trajectories

Agent Tool Use Trajectories (10K) 🚀

Dataset Description

This dataset contains 10,000 highly complex, multi-step dialogue trajectories designed to train open-source Large Language Models (LLMs) in advanced Agent Tool Use, Function Calling, and Reasoning.

Curated with professional AI training and data annotation standards, this dataset moves beyond simple synthetic Q&A pairs. It strictly follows the ChatML format and focuses heavily on multi-tool orchestration, logical step-by-step reasoning (Chain of Thought), and self-correction mechanisms.

• Total Rows: 10,000
• Format: ChatML (messages list)
• Supported Frameworks: Optimized for LangGraph, Agno, CrewAI, and custom Agentic Workflows.

Tool Scope & Capabilities

The trajectories simulate an autonomous agent interacting with a robust set of real-world tools, including but not limited to:

• Database Administration: Text-to-SQL (PostgreSQL execute_query) for complex joins and debugging.
• Software Engineering (SWE): Codebase vulnerability analysis and Git operations (create_pull_request, analyze_code_vulnerabilities).
• System & OS: Bash execution and file system manipulation (read_file, write_file, execute_bash_command).
• Web & RAG: Web scraping, search integration (e.g., Wikipedia/Tavily), and vector database retrieval.

Key Features

1. Strict Schema Adherence: The agent responses strictly match the JSON schemas provided in the system prompt without hallucinating non-existent parameters.
2. Self-Correction: A significant portion of the dataset includes scenarios where a tool returns an error (e.g., SyntaxError, FileNotFoundError). The agent is trained to read the error, re-evaluate its reasoning, and call the tool again with corrected parameters.
3. Sequential Reasoning: Enforces step-by-step tool execution rather than overwhelming parallel tool calls, allowing for true dependent reasoning (e.g., read file -> analyze content).

Data Structure

Each row in the dataset is a dictionary containing a single messages key, which holds the ChatML formatted conversation list:

{
  "messages": [
    {
      "role": "system",
      "content": "You have access to the following tools: [{'name': 'execute_postgres_query', ...}]"
    },
    {
      "role": "user",
      "content": "Find the highest salary in the IT department."
    },
    {
      "role": "assistant",
      "content": "I need to query the database to find the maximum salary.",
      "tool_calls": [
        {
          "name": "execute_postgres_query",
          "arguments": {"query": "SELECT MAX(salary) FROM employees WHERE dept = 'IT';"}
        }
      ]
    },
    {
      "role": "tool",
      "content": "{\"max_salary\": 95000}"
    },
    {
      "role": "assistant",
      "content": "The highest salary in the IT department is 95,000."
    }
  ]
}

Intended Use

This dataset is highly recommended for fine-tuning smaller parameter models (e.g., Llama-3-8B, Mistral-7B, Qwen) to grant them reliable function-calling capabilities and transform them into competent core engines for autonomous agent frameworks.