Mogzi

A Multi-model autonomous assistant right in your terminal or CI/CD pipeline and open to multiple API providers and AI models

Perform engineering research, full coding tasks, and multiple workflow with either natural language or shell scripts 👈

Please note that this project is still in the alpha stage and being actively developed. We welcome various conteribution from the community.

Features ✨

• Supports both interactive chat and non-tnteractive interface with streaming AI responses
• Currently supports the OpenAI-compatible APIs enabling numerous provides like Requesty.ai, Openrouter.ai, Supernova, Cerebras, Deepseek, and more.
  • For now, the best results are with models having excellect tool and instruction following like those form Anthropic, Google, and OpenAI models.
• Profile-based configuration for easy switching between providers and models
• Cross-platform support (Windows, MacOS, Linux)
• Control file system access via tool appproval (readonly or all)
• Chat history persistence with session management
• Pipe support for integrating with shell workflows
• MCP-Support (coming soon)

Examples 💡

# Start a rich interactive chat
mogzi

# Switch to using your profile named 'sonnet'
mogzi --profile sonnet

# Resume a previous conversation
mogzi chat --session "My Project Discussion"

# List available sessions (shows most recent 10 by default)
mogzi session list

# Get detailed information about a specific session
mogzi session info "My Project Discussion"


# Translate a README.md to portugese in one shot
cat README.md | mogzi run -p "Translate to portugese"

# Pipe content to continue an existing session
echo "Review this code change" | mogzi chat --session project-review

# (coming soon) Run a no non-interactive nworkflow prompt
mogzi run -p workflows/generate-release-notes.md

Prerequisites ✅

Before running the application, ensure you have:

• Access to at least one supported API provider
• Configuration file set up with your API providers and profiles

Setup 🛠️

1. Download the latest release following the steps below.

   For Windows:

   • Download mogzi.exe and moves it to your WindowsApp directory

   Start-BitsTransfer -Source https://github.com/peakflames/mogzi/releases/latest/download/mogzi-win-x64.exe -Destination mogzi.exe; move -Force mogzi.exe $env:USERPROFILE\\AppData\\Local\\Microsoft\\WindowsApps

   For MacOS (sudo):

   • Download mogzi and moves it to your /usr/local/bin

   sudo curl -L -o mogzi https://github.com/peakflames/mogzi/releases/latest/download/mogzi-osx-x64 && sudo chmod +x mogzi && sudo mv -f mogzi /usr/local/bin

   For Linux (sudo)

   • Download mogzi and moves it to your /usr/local/bin

   sudo curl -L -o mogzi https://github.com/peakflames/mogzi/releases/latest/download/mogzi-linux-x64 && sudo chmod +x mogzi && sudo mv -f mogzi /usr/local/bin

2. In your home directory, create a configuration file (mogzi.config.json) with your API provider details:

   {
       "mogziConfig": {
           "sessionListLimit": 10,
           "apiProviders": [
               {
                    "name": "MyCompanyProvider",
                    "type": "OpenAI-Compatible",
                    "apiKey": "example-key",
                    "baseUrl": "https://litellm.mycompany.com"
                },
                {
                    "name": "RequestyAI",
                    "type": "OpenAI-Compatible",
                    "apiKey": "example-key",
                    "baseUrl": "https://router.requesty.ai/v1"
                },
                {
                    "name": "Deepseek",
                    "type": "OpenAI-Compatible",
                    "apiKey": "example-key",
                    "baseUrl": "https://api.deepseek.com"
                }
           ],
           "profiles": [
               {
                   "default": true,
                   "name": "Default",
                   "apiProvider": "MyCompanyProvider",
                   "modelId": "03-mini"
               },
               {
                   "name": "Sonnet",
                   "apiProvider": "RequestyAI",
                   "modelId": "vertex/anthropic/claude-3-7-sonnet"
               },
               {
                    "name": "Gemini",
                    "apiProvider": "RequestyAI",
                    "modelId": "google/gemini-2.5-pro"
                },
                {
                    "name": "V3",
                    "apiProvider": "Deepseek",
                    "modelId": "deepseek-chat"
                }
           ]
       }
   }

Configuration Schema ⚙️

The application uses a JSON configuration file with the following structure:

• apiProviders: List of available API providers

  • name: Unique identifier for the provider
  • type: Provider type (OpenAI-Compatible at this time, more to come)
  • apiKey: Your API key for the provider
  • baseUrl: Base URL for the API (for OpenAI-Compatible providers)

• profiles: List of available profiles

  • default: Whether this is the default profile (true/false)
  • name: Profile name
  • apiProvider: Name of the API provider to use (must match a provider name)
  • modelId: Model ID to use for chat completion

Tool Approval 🔒

Mogzi includes a tool approval feature to provide control over file system operations. This setting can be configured in your mogzi.config.json file or overridden at runtime with a command-line argument.

Configuration:

To set the default tool approval mode, add the tool_approvals property to your mogzi.config.json:

{
    "mogziConfig": {
        "tool_approvals": "all",
        "apiProviders": [ ... ],
        "profiles": [ ... ]
    }
}

Command-Line Override:

You can override the configuration file setting using the --tool-approvals or -ta argument:

mogzi "Create a new file" -ta all

Modes:

• readonly: (Default) Mogzi will ask for your permission before performing any write operations (e.g., creating or modifying files).
• all: Mogzi is pre-approved to perform any file system operation without asking for confirmation.

Contributing 🤝

To contribute to the project, start with our Contributing Guide to learn the basics. You can also join our Discord to chat with other contributors in the #contributors channel.

Available Assistant Tools 🛠️

Mogzi is equipped with a powerful set of tools to interact with your local system. Here is a summary of the currently implemented tools:

Tool Name	Status	Notes	Safety/Security Features
list_directory	✅	Lists files and directories in a specified path. Supports glob patterns and .gitignore filtering.	Read-only operation. Constrained to the working directory. Respects .gitignore patterns by default.
read_text_file	✅	Reads the content of text files with optional line range support for large files.	Read-only operation. Constrained to the working directory. Supports pagination for large files.
read_image_file	✅	Reads and analyzes image files (PNG, JPEG, GIF, BMP, WebP).	Read-only operation. Constrained to the working directory. Returns base64-encoded image data for AI analysis.
read_pdf_file	✅	Extracts text content from PDF files.	Read-only operation. Constrained to the working directory. Extracts readable text from PDF documents.
write_file	✅	Creates a new file or overwrites an existing one with content.	- Requires --tool-approvals all. - Constrained to the working directory. - Uses atomic writes with verification. - Respects read-only file attributes.
replace_in_file	✅	Replaces specific text within files with precise context matching.	- Requires --tool-approvals all. - Constrained to the working directory. - Uses atomic operations with backups. - Requires exact text matching for safety.
apply_code_patch	✅	Applies Git-style unified diff patches for precise code changes.	- Requires --tool-approvals all. - Constrained to the working directory. - Uses fuzzy matching for robust patch application.
generate_code_patch	✅	Creates unified diff patches showing changes between content.	Read-only operation. Generates patches without modifying files.
preview_patch_application	✅	Previews what changes a patch would make without applying them.	Read-only operation. Safe preview of potential changes.
execute_command	✅	Executes shell commands with cross-platform support.	- Requires --tool-approvals all by default. - Cross-platform aware (uses cmd, zsh, bash appropriately). - Non-interactive commands only.
search_file_content	✅	Regex-based search across file contents with glob filtering.	Read-only operation. Constrained to the working directory. Supports pattern matching and file filtering.
attempt_completion	✅	Signals task completion and presents results to the user.	Read-only operation. Provides structured completion feedback and optional demonstration commands.

Tool Categories

File System Operations (Read-Only)

• list_directory - Browse directory contents
• read_text_file - Read text files with pagination support
• read_image_file - Read and analyze image files
• read_pdf_file - Extract text from PDF documents

File System Operations (Write - Requires --tool-approvals all)

• write_file - Create or overwrite files
• replace_in_file - Make targeted text replacements in files
• apply_code_patch - Apply unified diff patches

Code Analysis & Search

• search_file_content - Search file contents with regex patterns
• generate_code_patch - Create diff patches
• preview_patch_application - Preview patch changes

System Operations

• execute_command - Run shell commands (requires approval)
• attempt_completion - Signal task completion

Upcoming Tools

• list_code_definition_names - Extract code structure and definitions
• mcp_tools - Model Context Protocol support for external integrations

Copyright ©️

Copyright (c) 2025 Todd Schavey

This work is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc/4.0/ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.

Acknowledgements

• Cline
• Gemini CLI
• Microsoft.Extensions.AI
• Spectre.Console