muaddib 1.0.0

AI chatbot for IRC via irssi-varlink

🐁 Muaddib - a secure, multi-user AI assistant

Muaddib is an AI assistant that's been built from the ground up not as a private single-user assistant (such as the amazing Clawdbot / Moltbot), but as a resilient entity operating in an inherently untrusted public environment (public IRC servers).

What does it take to talk to many strangers?

1. It operates sandboxed, and with complete channel isolation.
2. It has been optimized for high cost and token efficiency (using a variety of context engineering etc. techniques).
3. It operates in "lurk" mode by default (rather than replying to everything, Muaddib replies when highlighted, but can also interject proactively when it seems useful).

Other work-in-progress features are also going to be tailored to this scenario (e.g. per-user token usage tracking and limiting / billing, per-channel code secrets and persistent workspaces, ...).

Of course, this means a tradeoff. Muaddib is not designed to sift through your email and manage your personal calendar! It is tailored for public and team environments, where it's useful to have an AI agent as a "virtual teammate" - both as an AI colleague in chat for public many-to-many collaboration, and allowing personal or per-channel contexts.

Quick Demo

Muaddib maintains a refreshing, very un-assistanty tone of voice that optimizes for short, curt responses (sometimes sarcastic, always informative) with great information density. And you may quickly find that Muaddib (in this case equipped with Opus 4.5) can do things that official Claude app does much worse (let alone other apps like ChatGPT or Gemini!).

➜ Generated image

(By the way, the token usage has been optimized since!)

Of course, as with any AI agent, the real magic is in chatting back and forth. (Multiple conversations with several people involved can go on simultaneously on a channel and Muaddib will keep track!)

(➜ Generated image, in case you are curious)

(Note that this particular task is on the edge of raw Opus 4.5 capability and all other harnesses and apps I tried failed it completely.)

Features

• AI Integrations: Anthropic Claude (Opus 4.5 recommended), OpenAI, DeepSeek, any OpenRouter model (including Gemini models), Perplexity AI
• Agentic Capability: Ability to visit websites, view images, perform deep research, execute Python code, publish artifacts
• Restartable and Persistent Memory: All state is persisted; AI agent maintains a continuous chronicle of events and experiences to refer to
• Command System: Automatic model routing (to balance cost, speed and intelligence) plus extensible command-based interaction with prefixes for various modes
• Proactive Interjecting: Channel-based whitelist system for automatic participation in relevant conversations
• [BETA] Long-running Projects: A quest mode (opt-in) that enables Muaddib to work on longer-horizon, many-step tasks in public, using the channel for long-term context and external steering

Muaddib has been battle-tested since July 2025 in a (slightly) hostile IRC environment, lurking at a variety of libera.chat channels. However, bugs are possible (no warranty etc.) and LLM usage carries some inherent risks (e.g. an E2B code execution sandbox with your API keys preloaded plus an access to the internet can be fooled by a highly crafted malicious website that the agent visits to upload these API keys somewhere).

Currently, only an IRC backend (using irssi connected via a varlink protocol) is implemented, but Discord and Slack backends are coming very soon.

Getting Started

Configuration

Copy config.json.example to config.json and set your:

• API keys (you can get started with just a small subset)
• Paths for tools and artifacts
• Custom prompts for various modes
• IRC integration settings such as channel modes

Installation

Recommended: See Docker instructions for running a Muaddib service + irssi in tandem in a Docker compose setup.

Manual ("bring your own irssi"):

1. Ensure irssi-varlink is loaded in your irssi
2. Install dependencies: uv sync --dev
3. Run the service: uv run muaddib

Commands

• mynick: message - Automatic mode
• mynick: !h - Show help and info about other modes

Development

# Install development dependencies
uv sync --dev

# Run tests
uv run pytest

# Run linting and formatting
uv run ruff check .
uv run ruff format .

# Type checking
uv run pyright

# Install pre-commit hooks
uv run pre-commit install

CLI Testing Mode

You can test the bot's message handling including command parsing from the command line:

uv run muaddib --message "!h"
uv run muaddib --message "tell me a joke"
uv run muaddib --message "!d tell me a joke"
uv run muaddib --message "!a summarize https://python.org" --config /path/to/config.json

This simulates full IRC message handling including command parsing and automatic mode classification, useful for testing your configuration and API keys without setting up the full IRC bot.

Chronicler

The Chronicler maintains persistent memory across conversations using a Chronicle (arcs → chapters → paragraphs) provided via a NLI-based subagent.

# Record information
uv run muaddib --chronicler "Record: Completed API migration" --arc "project-x"

# View current chapter
uv run muaddib --chronicler "Show me the current chapter" --arc "project-x"

Classifier Analysis

Evaluate the performance of the automatic mode classifier on historical data:

# Analyze classifier performance on database history
uv run python analyze_classifier.py --db chat_history.db

# Analyze classifier performance on IRC log files
uv run python analyze_classifier.py --logs ~/.irssi/logs/freenode/*.log

# Combine both sources with custom config
uv run python analyze_classifier.py --db chat_history.db --logs ~/.irssi/logs/ --config config.json

Results are saved to classifier_analysis.csv with detailed metrics and misclassification analysis.

Proactive Interjecting Analysis

Evaluate the performance of the proactive interjecting feature on historical data:

# Analyze proactive interjecting performance on database history
uv run python analyze_proactive.py --limit 20

# Analyze proactive interjecting on IRC log files with channel exclusions
uv run python analyze_proactive.py --logs ~/.irssi/logs/ --limit 50 --exclude-news

# Combine both sources with custom config
uv run python analyze_proactive.py --db chat_history.db --logs ~/.irssi/logs/ --config config.json

Results are saved to proactive_analysis.csv with detailed interjection decisions and reasoning.