__   __  ___  __       __  ___   __
 / /  / / /  / / /      / / /  /  / /
/ /__/ / /  / / /___   / / /  /  / /
/ ___  / /  / / ___  / / / /  /  / /
/ /  / / /  / / /  / / / / /  /  / /
/_/  /_/ /__/ /_/  /_/ /_/ /__/ /__/
        

From the void of computation, a new oracle speaks -- no gods, no cloud, only the terminal.

Fully local, offline AI in your terminal.
Zero API keys. Zero cloud. Zero limits.

Open-Mythos-2 runs GGUF models on your hardware via llama-cpp-python. Metal on macOS, Vulkan on Linux AMD, CUDA on NVIDIA. No data leaves your machine.

Get Started View Install Options
$ mythos model download
Downloading Mythos 4.8 (Gemma 3 12B fine-tune)...
$ mythos
Oracle ready. Type your question.
Fully Offline
Apache 2.0
Python 3.10+
Vulkan / Metal / CUDA
Scroll

Why Mythos

Cloud AI services trade your privacy for convenience. Mythos gives you both.

Feature Mythos Cloud AI
Data privacy 100% local -- nothing leaves your machine Data sent to remote servers
API keys required None Required, metered
Offline use Fully offline Requires internet
Cost Free forever Per-token billing
Security scanning Built-in SAST + AI audit Not included
Self-improvement RML feedback engine Static behavior
License Apache 2.0 Proprietary

Features

Everything you need for local AI, in one package.

>_

Terminal Chat

Rich terminal UI with syntax highlighting, streaming responses, slash commands, and full conversation management. Your AI lives in the terminal where you do.

/

Slash Commands

Built-in commands for everything: /config, /export, /summary, /voice, /rag, /fix, /rml, /persona, and more. Full control without leaving the chat.

[*]

RML Feedback Engine

Reinforcement Meta-Learning: Mythos learns from your feedback. Explicit signals (/rml good/bad) and implicit cues adjust temperature, top_p, and system prompts over time.

{}

Security Scanner

Static SAST scanner for quick findings, AI-powered deep scanner for nuanced audits. Auto-fix with --apply or inline /fix in chat. No external tools needed.

[R]

RAG Pipeline

ChromaDB + sentence-transformers. Index local documents, query them in chat. Perfect for codebases, docs, or research papers -- all local, all offline.

~~

Voice I/O

Whisper-based speech-to-text input. espeak-ng or Piper text-to-speech output. Press V to talk, let Mythos talk back. Hands-free local AI.

G

Gradio Web UI

Not a terminal person? Launch the web interface with --mode web. Same engine, same model, browser-based interaction on port 7860.

^

LoRA Fine-Tuning

Train custom LoRA adapters on your own data. Merge into the base model, convert to GGUF, and run your fine-tuned model locally. Full pipeline included.

::

Cross-Session Memory

Key facts persist across sessions. Session summaries give structured digests. Your AI remembers context without sending anything anywhere.

RML: Reinforcement Meta-Learning

The only local AI that improves the more you use it.

01

You chat and give feedback

Use /rml good or /rml bad to explicitly rate responses. Mythos also picks up implicit signals: edits, interruptions, and keyword cues.

02

Scores accumulate per category

Feedback is categorized (reasoning, creativity, instruction-following, etc.) and scored. When a category crosses the negative threshold, an adjustment triggers.

03

Generation params are nudged

Temperature, top_p, and other sampling parameters are automatically adjusted. A learned hint is injected into the system prompt to steer future responses.

04

Preferences persist

RML state is saved to ~/.config/mythos/rml_preferences.json. Your AI gets better across sessions, not just within one conversation.

Security Scanner

Built-in SAST and AI-powered security auditing. No external tools.

Static Scan

Regex-based SAST
No model needed

-->

Deep Scan

AI-powered audit
Context-aware

-->

Auto-Fix

Dry-run or apply
MYTHOS_PATCH rewrite

-->

Verify

Re-scan after fix
Confirm resolution

$ mythos scan ./my-project --deep

[SCAN] Scanning 47 files...
[HIGH] SQL injection in auth.py:42 -- raw string interpolation in query
[MED] Path traversal in upload.py:18 -- unsanitized filename in open()
[LOW] Hardcoded debug flag in config.py:5

$ mythos fix --path ./my-project --apply

[FIXED] auth.py:42 -- parameterized query
[FIXED] upload.py:18 -- path validation with Path.is_relative_to()
[SKIP] config.py:5 -- requires manual review

Get Started

Up and running in three commands.

1

Install Mythos

# Install from PyPI
pip install mythos-sentinel
2

Download a model

# Downloads Mythos 4.8 (Gemma 3 12B fine-tune, Q4_K_M)
mythos model download
3

Start chatting

mythos
*

Or run from source

git clone https://github.com/creatorofsomethingthatisgood/Open-Mythos-2.git
cd Open-Mythos-2
python main.py --mode chat

Command Reference

The mythos CLI at a glance.

Command Description
mythosLaunch terminal chat (default mode)
mythos cloudCloud API mode (OpenAI-compatible)
mythos scan <path>Static security scan
mythos scan <path> --deepAI-powered deep security audit
mythos fix --path <path>Dry-run auto-fix for findings
mythos fix --path <path> --applyApply auto-fixes to disk
mythos model downloadDownload default GGUF model
mythos doctorDiagnose setup issues
mythos configView or edit configuration
mythos historyList past conversation sessions

Open Source

Public domain. No restrictions. Build anything.

0
API Keys Required
100%
Local
0
Data Sent to Cloud
Apache 2.0 License

Mythos is released under the Apache License 2.0.
Use it, fork it, sell it, embed it -- permissive and powerful.

Star on GitHub Report an Issue