v0.3.0

tool-maker v0.3.0

What's Changed

• refactor: introduce multi-backend database support with SQLite and Po… by @codewithwest in #1

New Contributors

• @codewithwest made their first contribution in #1

Full Changelog: v0.2.0...v0.3.0

v0.2.0

tool-maker v0.2.0

Full Changelog: v0.1.0...v0.2.0

v0.1.0

v0.1.0 — Initial Release

Date: 2026-06-28

The first public release of LLM Tool Maker — an AI-powered tool generation and execution system that uses local LLMs to create, test, and run Python tools from natural-language descriptions.

Features

Core Orchestration

• ToolMaker class that wires together analysis, generation, and execution
• ToolMakerConfig dataclass with provider, model, and workspace settings
• Factory functions: create_tool_maker() and create_tool_maker_from_env()

Project Analysis

• ProjectScanner with AST-based parsing of Python source files
• Extracts functions, classes, imports, and entry points
• Reads pyproject.toml and requirements.txt for dependency detection
• JSON export and human-readable summaries

LLM Integration

• OllamaProvider — communicates with local Ollama server via HTTPX
• Auto model resolution with fallback to first available model
• Conversation history support
• Detailed request/response logging

Tool Generation

• ToolGenerator creates Python tools from natural-language descriptions
• Code extraction from LLM responses with regex parsing
• Function signature and dependency detection
• Fallback stub generation when no LLM is available
• Project-aware generation that includes context from scanned projects

Tool Execution

• ToolExecutor runs generated tools with result capture
• File-based and multi-tool batch execution
• Execution history and in-memory caching via ToolManager

Sandbox Security

• Subprocess-based sandbox with python -I isolated mode
• 60+ whitelisted standard library modules
• 30+ safe built-in functions
• Custom __import__ hook blocks non-whitelisted modules
• Configurable timeout with subprocess termination
• Auto-serialization of return values to JSON-safe types

CLI

• analyze <path> — scan a Python project
• generate <query> — generate a tool from a description
• run <file> — execute a saved tool
• ui — launch the web dashboard
• version — print version information

Web UI (Flask)

• Dashboard with live log streaming via SSE
• Project analysis page
• Tool execution sandbox
• LLM provider testing interface
• Configuration management

Flask Extension

• ToolMakerExtension for easy integration into Flask apps
• Blueprinted health check, analyze, and tool endpoints

Configuration

• ToolMakerConfigFile persisted to ~/.config/tool-maker/config.json
• dotenv.py — zero-dependency .env loader

Testing

• 9 test files covering core components
• Sandbox timeout and error handling tests

---

Contributors: Jonas