slack-tone-bot

I built this out of pure laziness — I was tired of formatting my messages manually every single time, so I automated it into my slack workspace ;). (Will be working on context of previous chats in a session soon.)

Local Slack bot that reformats messages into your personal tone, running entirely on-device using MLX and a quantized Mistral 7B model.

---

Features

• Personal tone rewriter: Takes any Slack message and rewrites it to match your configured style.
• Two entry points:
  • /fmt slash command (ephemeral response).
  • DM the bot directly and it replies with a reformatted version.
• Local-only LLM: Uses mlx-community/Mistral-7B-Instruct-v0.3-4bit via mlx-lm.
• Configurable style: Tune persona, tone, and examples in config/style.yaml.

---

Project structure

• src/slack_tone_bot/ — bot code (LLM wrapper, formatter, Slack Bolt app, entrypoint).
• config/style.yaml — tone/style configuration (persona, sliders, guidelines, examples).
• .env — Slack credentials and optional overrides (not committed; use .env.example as a template).

Entrypoint:

• CLI script: slack-tone-bot (via pyproject.toml).
• Module: python -m slack_tone_bot.main

---

Requirements

• Python 3.11+
• macOS with Apple Silicon recommended (for MLX performance)
• Slack app with:
  • Bot token (SLACK_BOT_TOKEN)
  • App-level token (SLACK_APP_TOKEN, with connections:write for Socket Mode)

Python dependencies are managed via pyproject.toml and include:

• slack-bolt, slack-sdk
• mlx-lm
• pydantic
• PyYAML
• python-dotenv

---

Setup

1. Clone and create a virtualenv

git clone <your-repo-url> slack-tone-bot
cd slack-tone-bot

python -m venv .venv
source .venv/bin/activate

2. Install in editable mode

pip install --upgrade pip
pip install -e .

For src/ layout projects like this one, using pip install -e . ensures slack_tone_bot is importable without manually tweaking PYTHONPATH.

3. Configure environment

Create a .env file in the project root (you can copy .env.example):

cp .env.example .env

Then fill in at least:

SLACK_BOT_TOKEN=xoxb-...
SLACK_APP_TOKEN=xapp-...

Optional overrides:

• MODEL_NAME — MLX model name (default: mlx-community/Mistral-7B-Instruct-v0.3-4bit)
• STYLE_CONFIG_PATH — path to a custom style YAML (default: config/style.yaml)

4. Customize the tone (optional but recommended)

Edit config/style.yaml to match your personal style:

• persona — short description of how you communicate.
• tone — sliders for formality, warmth, humor, energy, emoji usage, capitalization, sentence style.
• guidelines — bullet rules like “keep messages short”, “avoid jargon”.
• examples — input/output pairs the model should mimic.

---

Running the bot

Activate your virtualenv (if not already):

source .venv/bin/activate

Then start the bot:

python -m slack_tone_bot.main

On first run, MLX will download the model weights and initialize the tokenizer, which can take 10–20 seconds.

If the Slack tokens or style config path are missing/invalid, the process will log an error and exit.

---

Using the bot in Slack

• Slash command: In any channel or DM where the bot is installed, run:

  /fmt I think we should maybe delay the deploy a bit until we understand the metrics better
  

  The bot replies ephemerally with a reformatted version in your tone.

• Direct messages: DM the bot any text, and it will respond with a rewritten version:

  hey can we sync on this later this week
  

  → reformatted reply in your configured style.

---

Implementation notes

• The system prompt is built from config/style.yaml via build_system_prompt and combined with the user message before sending to the LLM.
• Inference is done through a small LocalLLM wrapper (mlx_lm.load + mlx_lm.generate) with a per-instance lock for basic thread safety.
• The Slack app runs in Socket Mode via slack_bolt’s SocketModeHandler, with:
  • /fmt command for explicit rewrites.
  • message event handler for DMs.

---

Troubleshooting

• Imports fail (ModuleNotFoundError: slack_tone_bot)

  • Ensure you ran pip install -e . inside the virtualenv.
  • Or, as a quick check: PYTHONPATH=src python -c "from slack_tone_bot.config import load_style_config; print('ok')"

• Bot exits with “Missing SLACK_BOT_TOKEN or SLACK_APP_TOKEN”

  • Verify .env exists at the project root and the variables are set.

• Slow first run

  • Model download and initialization are expected to be slow the first time. Subsequent runs should be faster.

---

Development tips

• Activate the venv and run:

  source .venv/bin/activate
  python -m slack_tone_bot.main

• To quickly verify that imports and config loading work:

  python -c "from slack_tone_bot.config import load_style_config; print('ok')"