LLM Tool Maker

An intelligent tool-making system that uses LLMs to analyze projects, generate tools, and execute them in a sandboxed environment — all through a modern web dashboard or CLI.

Quick Start

# 1. Install
pip install llm-tool-maker

# 2. Initialize (detects Ollama, creates DB, writes .env)
llm-tool-maker init

# 3. Launch the dashboard
llm-tool-maker ui

That's it. No PostgreSQL, no external services — just your local Ollama instance. The dashboard opens at http://localhost:5000.

Features

Feature	Description
Zero-setup persistence	SQLite by default (stdlib, no dependencies). Opt-in PostgreSQL for production.
Pluggable LLMs	Ollama (local), OpenAI, or Anthropic — swap via --provider
Sandboxed execution	Subprocess runner with module whitelist, no network, configurable timeout, auto-retry with dep install
Autonomous pipeline	6-stage DB-backed pipeline: Analyse → Plan → Validate → Implement → Test → Review, with auto-fix loop
Web dashboard	6-page glass-morphism UI (Dashboard, Pipeline, Execute, Analyze, Provider, Config, Docs)
Dependency management	AST-based import scanning, 200+ stdlib modules, 50+ module→package mappings, auto-install
Remote API client	ToolMakerClient lets you consume Tool Maker as a REST service
Docker Compose	One-command deployment with Ollama + PostgreSQL + the app

Installation

# From PyPI
pip install llm-tool-maker

# With PostgreSQL support (optional)
pip install 'llm-tool-maker[postgres]'

# All extras
pip install 'llm-tool-maker[all]'

# From source
git clone https://github.com/codewithwest/project_tool-maker.git
cd project_tool-maker
uv sync

Usage

CLI

llm-tool-maker init            # One-time setup (checks Ollama, creates DB, writes .env)
llm-tool-maker ui              # Launch web dashboard
llm-tool-maker analyze <path>  # Scan a project
llm-tool-maker pipeline <goal> # Run full autonomous pipeline
llm-tool-maker run <file>      # Execute a tool file
llm-tool-maker config show     # View configuration
llm-tool-maker migrate up      # Run DB migrations
llm-tool-maker --help          # All commands

Python API

from tool_maker import ToolMaker

# Use local Ollama (default)
tm = ToolMaker(llm_provider="ollama", model="llama3.2")

# Or OpenAI
tm = ToolMaker(llm_provider="openai", api_key="sk-...", model="gpt-4o-mini")

# Or Anthropic
tm = ToolMaker(llm_provider="anthropic", api_key="sk-...", model="claude-sonnet-4-20250514")

# Analyze, generate, execute
info = tm.analyze_project("/path/to/project")
result = tm.create_and_execute_tool("Parse CSV files and return row count")

Remote API Client

Use Tool Maker as a remote service from any Python project:

from tool_maker import ToolMakerClient

client = ToolMakerClient("http://localhost:5000")
tools = client.list_tools()
result = client.execute("print('hello world')")
client.run_pipeline("Build a CLI tool that counts lines of code")

API-only Mode

Serve just the REST API (no Jinja templates):

llm-tool-maker ui --api-only

Database

Zero-config: SQLite is used automatically when no TOOLMAKER_DB_DSN is set. The database file lives at ~/.config/tool-maker/data.db.

PostgreSQL (for production):

export TOOLMAKER_DB_DSN="postgresql://user:pass@localhost:5432/toolmaker"
llm-tool-maker ui

Migrations run automatically on startup.

Docker

docker compose up

This starts Ollama, PostgreSQL, and the app — reachable at http://localhost:5000.

Environment Variables

Variable	Default	Description
TOOLMAKER_DB_DSN	"" (SQLite)	PostgreSQL DSN. Empty = SQLite backend.
TOOLMAKER_DB_PATH	~/.config/tool-maker/data.db	SQLite database file path
OLLAMA_BASE_URL	http://localhost:11434	Ollama server URL
OLLAMA_MODEL	llama3.2	Default LLM model
SANDBOX_TIMEOUT	30	Tool execution timeout in seconds
MAX_FIX_ATTEMPTS	3	Auto-fix loop retry limit
TOOL_MAKER_CONFIG	~/.config/tool-maker/config.toml	Config file path

Web Dashboard

Pages

• Dashboard — overview, live terminal, quick stats
• Pipeline — run the 6-stage autonomous pipeline with progress tracking
• Execute — IDE-style editor with sidebar (Saved + Database tools), command bar, tabbed results
• Analyze — scan a project and inspect its structure
• Provider — configure LLM provider and test prompts
• Config — manage database, migrations, sandbox whitelist, dependency approvals
• Docs — view release notes and README, rendered as formatted markdown

Project Structure

src/tool_maker/
├── __init__.py          # Public API exports
├── client.py            # ToolMakerClient (remote HTTP client)
├── config.py            # ToolMakerConfigFile
├── tool_maker.py        # Main orchestrator
├── tool_fixer.py        # LLM-driven tool fixer
├── analyzer/            # AST-based project scanner
├── cli/                 # CLI argument parsing and handlers
├── db/                  # SQLite + PostgreSQL backends, models, migrations
├── llm/                 # Ollama, OpenAI, Anthropic providers
├── planner/             # Planner, validator, executor, reviewer
├── tool/                # Generator, executor, sandbox, fixer, deps
└── ui/                  # Flask web app (routes, templates, static)

Development

uv sync
uv run ruff check .
uv run pytest

License

MIT