Task Orchestrator via Local LLM (Ollama / LM Studio)

Splits a task into subtasks and executes them step-by-step via a local LLM through Ollama or LM Studio.

Requirements

• Python 3.14+
• No third-party packages — script.py uses only the Python standard library (urllib)
• A running Ollama (http://localhost:11434) or LM Studio (http://localhost:1234) instance with at least one model pulled/loaded

Installation

No installation required — just run start.bat (Windows), it creates the venv/ virtual environment automatically on first launch. You can also run directly with your system Python:

python script.py

Usage

1. Write your overall task description into task.md (plain Markdown, any language).
2. Run the script:

python script.py

On the first run you will be asked to choose a backend (Ollama / LM Studio) and a model. The choice is saved to config.json.

CLI flags:

• --reconfigure — ask for backend/model again and overwrite config.json

  python script.py --reconfigure

• --resume — continue from the last unfinished step (based on progress.json), without re-planning

  python script.py --resume

• --dry-run — only generate the subtask plan into tasks/, do not execute them

  python script.py --dry-run

• --lang {en,ru} — language of the generated plan/subtasks (default: en); console messages stay in English

  python script.py --dry-run --lang ru

On Windows you can also use start.bat (it picks venv/ Python if present and forwards all arguments):

start.bat --dry-run

Task Orchestrator via Local LLM