Miki 1.1.0 - Mini Claude Code CLI

Miki is a simple, Node.js-based CLI that connects to a local Ollama LLM to help you analyze, search, and edit files in your current directory.

Features

• Connects to Ollama: Works with your local LLM models (optimized for Llama 3.2).
• File System Operations: List, read, search (grep-like), and precisely edit files.
• Task Planning: Automatically breaks down your requests into small, actionable steps in a local plan.md file.
• Progress Tracking: Marks steps as complete as it performs them.
• Interactive Workflow: Proposes a plan, allows natural language modifications, and performs self-review after execution.
• Resumption: Continue interrupted tasks easily with the continue command.
• Reporting: Generates task summary reports in ./miki-reports/.
• Configurable: Master config file for Ollama URL, model, and auto-proceed settings.

Installation

npm install -g miki-cli

Usage

Run the tool using the miki command:

miki

Commands inside the CLI:

• [Request]: Enter any task (e.g., "Implement a login form in React").
• continue: Resumes the task from the first incomplete step in plan.md.
• config: Change Ollama URL, model, or auto-proceed settings.
• exit or quit: Leave the CLI.

How it works (Miki 1.1.0 Workflow)

1. Clarification: Miki analyzes your request and asks for clarification if needed.
2. Planning: Miki generates a Markdown task list (plan.md) in your current directory.
3. Confirmation: You review the plan and can choose to:
   • Proceed: Start execution immediately.
   • Modify: Provide natural language feedback to update the plan.
   • Cancel: Stop the task.
4. Execution: Miki uses tools to perform each step, automatically marking them as [x] in plan.md when done.
5. Self-Review: Miki reviews its own changes to ensure quality.
6. Report: A summary is saved to ./miki-reports/.

Best Practices

• Review the Plan: Always glance at the proposed plan.md. If it's too broad, use the Modify option to ask for more granular steps.
• Keep Steps Small: Miki performs best when steps are clear and focused on a single file or logic block.
• Resuming with continue: If Miki stops unexpectedly or you hit a token limit, just type continue to pick up exactly where it left off.
• Manual Plan Edits: You can manually edit plan.md between requests if you want to skip steps or re-order them. Miki respects your changes!
• Model Choice: While Miki works with many models, llama3.2:8b or higher is recommended for the most reliable tool use and planning.

Configuration

Settings are stored in ~/.miki/config.json. You can also change them by typing config inside Miki.

• ollama_url: The address of your Ollama server (default: http://localhost:11434).
• model: The model to use (default: llama3).
• auto_proceed: If true, Miki skips the analysis confirmation, but will always ask for plan confirmation in 1.1.0.
• shell_auto_proceed: If true, Miki will execute shell commands without asking.