MindLoc CLI

AI-powered terminal application for chatting and coding using local models via Ollama.

Note: This package was previously published as localmind. I've renamed it to mindloc after becoming aware of another application with the same name.

Important: MindLoc is a CLI tool and must be installed globally with the -g flag.

Installation

npm install -g mindloc-cli

After installation, run from any terminal:

mindloc

Prerequisites

• Node.js (v18 or higher)
• Ollama installed and running

Features

Dual Mode System

• Chat Mode - Conversational AI with web search and deep research capabilities
• Code Mode - Secure project-scoped development environment with intelligent coding assistance

Intelligent Context Management

• Automatic context condensing at 80% capacity
• Preserves recent messages and system prompts
• Condenses older conversation into concise summaries
• Per-model conversation memory that persists across sessions

Auto Code Mode

• AI can automatically execute actions with intelligent approval system
• Automatic action detection from AI responses
• Smart approval workflow (immediate, pre-approval, or post-approval based on action type)
• Automatic rollback on rejection

Hybrid RAG (Retrieval-Augmented Generation)

• Combine local knowledge base with web search
• Store code snippets and documentation locally
• Search and retrieve relevant information automatically
• Available in both chat and code modes

Advanced Features

• Syntax Highlighting - Beautiful code blocks with language detection
• Live Diff Viewer - Side-by-side comparison of file changes with color-coded additions/deletions
• Session Dependency Graph - Track and visualize session provenance
• Thinking Mode - See model's reasoning process in real-time (for supported models)
• Web Search - Integrated DuckDuckGo search for up-to-date information
• Resource Monitoring - Real-time CPU, RAM, GPU, and VRAM usage display
• Interactive Settings - Configure all features without editing config files
• Theme Support - 5 built-in themes (Default, Ocean, Forest, Sunset, Monochrome)

Security

• Mandatory Project Directory - All file operations scoped to specified directory in code mode
• Path Validation - Prevents AI from accessing files outside project scope
• Command Context - Shell commands execute within project directory

Quick Start

Interactive Mode

mindloc

Select your mode (Chat or Code) and model, then start chatting!

Direct Mode Launch

# Start chat mode
mindloc chat

# Start code mode
mindloc code

# Use specific model
mindloc chat -m llama3.2
mindloc code -m codellama

# Quick question
mindloc ask "How do I implement a binary search in Python?"

# List available models
mindloc models

# Get help
mindloc help
mindloc help chat

Chat Mode Commands

• /settings - Configure MindLoc settings
• /think [off] - Toggle/configure thinking mode (for supported models)
• /search <query> - Web search
• /research <topic> - Deep research with multiple searches
• /rag add - Add snippet to local knowledge base
• /rag list - List all stored snippets
• /rag search <query> - Search local knowledge
• /graph - Show session dependency graph
• /why - Explain AI's reasoning for last answer
• /save [title] - Save conversation
• /load - Load previous conversation
• /new - Start new conversation
• /ollama <model> - Pull a model from Ollama
• /unload - Unload current model and switch
• /exit - Exit chat mode

Code Mode Commands

• /settings - Configure MindLoc settings
• /think [off] - Toggle/configure thinking mode (for supported models)
• /file <path> - Load file into context
• /write <path> - Write AI response to file
• /exec <command> - Execute shell command
• /delete <path> - Delete a file
• /project [path] - View/set project directory
• /context - Show context usage
• /rag [add|list|search] - Knowledge base operations
• /graph - Show session dependency graph
• /save [title] - Save session
• /load - Load session
• /new - New session
• /exit - Exit code mode

Configuration

MindLoc stores data in ~/.mindloc/:

• config.json - Settings
• memory/ - Conversation history
• rag/ - Local knowledge base

Settings

Access settings with /settings command in any mode:

• Auto Code Mode - Enable/disable automatic action execution
• Hybrid RAG - Enable/disable local knowledge base
• Live Diff Viewer - Show file changes before approval
• Session Graph - Track session dependencies
• Theme - Choose from 5 built-in themes
• Global Context Length - Set default context window for all models

Supported Models

MindLoc works with any Ollama model. Popular choices:

• Chat: llama3.2, mistral, gemma, qwen
• Code: codellama, deepseek-coder, starcoder
• Thinking: qwen3, deepseek-r1, glm-4.7-flash, gpt-oss

Install models with:

ollama pull llama3.2
ollama pull codellama

Thinking Mode

Supported models with thinking capabilities:

• qwen3, deepseek-r1, deepseek-v3.1
• gpt-oss, gpt-oss-safeguard
• glm-4.7-flash, qwen3-vl, qwen3-next
• nemotron-3-nano, magistral

Thinking mode shows the model's reasoning process in real-time before providing the final answer.

Uninstall

npm uninstall -g mindloc-cli

License

MIT