Pepper

Your Personal AI Assistant (Blog)

Pepper is a personal AI assistant that proactively works for you. It connects to your Gmail to autonomously summarize important emails and surface critical updates that need your attention, delegating complex tasks to a swarm of workers that handle them seamlessly in the background.

Setup

First clone the repo:

git clone --recurse-submodules https://github.com/agentica-org/pepper

Prerequisites: Install and run the Context Store locally before launching Pepper.

conda create -n pepper python=3.12 pip -y
conda activate pepper

# First, install our context store Episodic
cd episodic-sdk
pip install -e .[semantic]

# Then, install requirements for Pepper
cd ../pepper
pip install -r requirements.txt

followed by:

episodic serve --port 8000

Keep the process running, and open a new terminal for the following setup.

Configure Environment

1. Copy the environment template:

   cd pepper
   cp env_var.example.sh env_var.sh

2. Set up Composio (required):

   • Sign in at composio.dev
   • Create an Project API key
   • Fill in it as COMPOSIO_API_KEY

3. Fill in your LLM API key (required):

   • For OpenAI (default): Set OPENAI_API_KEY
   • For Anthropic: Set ANTHROPIC_API_KEY

4. Configure LLM provider (optional):

   • By default, Pepper uses OpenAI models
   • To use Anthropic models:
     • Add to your env_var.sh: export LLM_PROVIDER="anthropic"
     • Update the MODEL variable in pepper/agent/scheduler.py, pepper/agent/worker.py, and pepper/agent/workflow.py
     • Example: MODEL = "claude-sonnet-4-5-20250929"

5. Load environment variables:

   source env_var.sh

6. Login your Gmail account and grant access [Only for the first time]:

   cd ~/pepper
   python -m pepper.services.email_service

   If you see the message "Please authorize Gmail by visiting: url", open the url in your browser and grant access.

   Once you've see the message "✅ Trigger subscribed successfully.", Ctrl+C to stop the process, you're good to go!

Running Pepper

Launch pepper (The first time you run this, it'll take a while to build your profile, approximately 1 minute):

python -m pepper.launch_pepper

Open the UI in your browser as prompted:

http://localhost:5050/pepper/ui.html

If you're in the remote server, vscode should be able to port forward the correct port to your local machine automatically.

Note: Press Ctrl+C to stop all services.

Acknowledgement

This work is done with the Agentica Team as part of Berkeley Sky Computing Lab.