slack-drafter

A lightweight GPT-powered Slack bot that helps you reply faster — without sounding like a robot.

Powered by GPT-4 + LangChain. Designed for thoughtful humans who want to save time.

---

Features

• ⌨️ Use /draft [thread URL] in Slack to generate a reply suggestion
• 🧵 Understands full thread context to avoid repetition
• 🔣 Adapts to your voice (IC-focused tone by default)
• 🔒 Replies are private until you choose to send them
• ⚡ Built with FastAPI, Slack Bolt, and LangChain

---

Quickstart

git clone https://github.com/mlane/slack-drafter.git
cd slack-drafter

# Create and activate virtual environment
python3.11 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Copy and edit environment variables
cp .env.sample .env
# Then add your OpenAI and Slack credentials

Run the app:

python src/main.py

---

Environment Variables

LANGCHAIN_API_KEY=your-langchain-api-key
LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
LANGSMITH_PROJECT=your-project-name
LANGCHAIN_TRACING_V2=true
OPENAI_API_KEY=your-openai-key
SLACK_ALLOWED_USERS=user-id-1,user-id-2
SLACK_BOT_TOKEN=xoxb-your-slack-token
SLACK_SIGNING_SECRET=your-signing-secret

Note: The LangChain and LangSmith variables are optional and currently unused. They're included for future support (e.g. memory, tracing, or LangSmith debugging).

---

Minimal Testing

To keep LLM costs low, we recommend basic tests for:

• extract_slack_ids()
• format_timestamp()
• format_user()

Example:

# tests/slack_utils_tests.py
def test_format_timestamp():
    ts = "1714774008.000000"
    formatted = format_timestamp(ts)
    assert formatted == "2024-05-03 10:06 PM UTC"

---

Slack App Setup

1. Create a Slack App at https://api.slack.com/apps
2. Enable Slash Commands and Event Subscriptions
3. Add these OAuth scopes:
   • commands
   • chat:write
   • channels:history
   • users:read
4. Create a slash command:
   • Command: /draft
   • Request URL: https://<your-domain>/slack/events
5. Install to your workspace

---

Roadmap

• Slash command reply suggester
• Ephemeral message previews
• Tone tuned to IC voice
• Optional App Home fallback UI
• Long-term personalization via memory/context
• Message quality scoring (e.g., “Too wordy”)

---

🧑‍💻 License

MIT

Contributions

PRs welcome. Please keep things minimal, tested, and low-dependency.

---

Why "slack-drafter"?

The goal is simple: help you draft thoughtful replies — fast. Whether you're in back-to-back meetings or catching up on threads, this Slack bot gives you a head start without losing your voice.

Previously named slack-reply-assistant.