!users will need personal API key for the choosen provider!
Changelog:
- Theme System: Introduced
/themecommand with 8 premium styles (hl3, matrix, cyberpunk, etc.). Make MARK unique! - Dynamic Splashes: improved startup experience with random splash messages!
- Local Models: Support for local LLMs via Ollama integration.
- Logo Refinement: Adjusted the terminal logo for a perfect 50/50 color split.
- Memory Improvement: Improved context awareness for personal memory.
- Stats Toggle: Added
/tscommand to show/hide session stats.
Mark is a personal, persistent, and intelligent assistant that lives entirely in your terminal. It combines the power of state-of-the-art LLMs (Google Gemini & Groq) with a local, secure memory system.
The coolest feature? Local Memory. If you tell Mark something, he remembers it forever (saved locally on your machine).
> remember that my preferred language is Python
✅ I will remember: "my preferred language is Python"
> write a script to sort a list
(Mark writes the script in Python automatically because he remembers!)
Mark's memory is 100% local. No data brokers, no cloud storage of your personal facts.
Kleos is Mark's advanced capabilities engine. It transforms Mark from a simple chatbot into a reasoning agent.
Why use /kleos?
Standard LLMs often rush to answer, missing context or nuance. Kleos forces a "Think before you speak" workflow.
- Analyze: You provide a prompt. Kleos analyzes it for ambiguity, missing context, or hidden requirements.
- Refine: It acts as an expert "Prompt Engineer", rewriting your query to be logically perfect.
- Execute: It sends the optimized prompt to the AI.
Result: Drastically higher quality answers, better code, and deeper insights.
Usage:
/kleos Explain how to build a neural network from scratch(Mark will pause, analyze your request, optimize the internal prompt, and then deliver a superior tutorial)
Requirements: Python 3.8+
# Clone the repository
git clone https://github.com/piskevalee-cpu/MARK.git
cd mark
# Install globally
pip install .To run Mark:
markOn first run, Mark guides you through setup.
Supported Providers (more to come..):
- Groq (Fast Inference, suggested)
groq/compoundllama-3.3-70b-versatileopenai/gpt-oss-120bmoonshotai/Kimi-K2-Instruct
- GOOGLE (BETA)
gemini-2.5-flashgemini-2.5-flash-litegemini-3-flash-previewgemma-3-27b
- LOCAL (via Ollama, BETA)
- Supports any model installed on your local Ollama instance (e.g.,
llama3.2,mistral,phi3). No API key required, just need to have ollama running!
- Supports any model installed on your local Ollama instance (e.g.,
You can change settings anytime with:
/configor switch models on the fly:
/model[NOTE: Mark has only been tested with free APIs, results may vary with paid APIs.] [NOTE2: Gemini implementation is still in beta, results may vary. We suggest using Groq for now.]
| Command | Description |
|---|---|
/help |
Show this help menu |
/kleos <prompt> |
Activate Kleos System (Reasoning Mode) |
/web <prompt> |
Search the Web (Real-time info) |
/model |
Switch AI Provider & Model |
/theme <name> |
Change UI Theme (hl3, matrix, etc.) |
/config |
Update API Keys and Defaults |
/memory list |
View all saved memories |
/memory search <q> |
Semantic search in memory |
/remember <text> |
Save a fact to local memory |
/ts |
Toggle session statistics visibility |
/reset |
Clear current conversation context |
/exit |
Close Mark |
mark/
├── mark_cli/ # Core package
│ ├── main.py # Application entry point
│ ├── memory/ # SQLite vector memory system
│ ├── providers/ # Async AI implementations
│ └── ui/ # Rich-based TUI implementation
├── docs/ # Website & Documentation (coming soon)
└── scripts/ # Verification tools
API keys are stored securely using your OS's native keychain (Windows Credential Manager, macOS Keychain, or Linux Secret Service). They are never stored in plain text.
Created by Valerio Pischedda (piskevalee-cpu). Started as a simple Telegram bot for school schedules, evolved into a powerful terminal AI companion.
Note that this tool has been partially vibecoded and debugged by AI, it is still in beta, and may have some bugs. Please report any issues you find on the GitHub repository, or send me an e-mail at piskevalee@gmail.com !
MIT License