Bub

Bub is a common shape for agents.

It exists to answer a harder question than "can an agent finish a task?": when many humans and agents work in the same environment, what kind of agent remains understandable, reviewable, and safe to continue?

Bub's answer is an agent form with explicit boundaries, visible evidence, and safe handoff. The current repository is one implementation of that idea, using hook-based composition, Republic as the context runtime, and constructing context from tape as the current context model.

Quick Start

Install dependencies and create local config:

git clone https://github.com/bubbuild/bub.git
cd bub
uv sync
cp env.example .env

Run interactive local chat:

uv run bub chat

Run a one-shot task:

uv run bub run "summarize this repository"

Start channel listener mode:

uv run bub gateway

Deployment

For production setup and operations, read:

• Deployment Guide
• Channels Overview
• Telegram Channel

Read Next

• Core Overview: architecture and capability summary in one place
• Workflows Overview: CLI and skills usage in one place
• Extension Guide: build and publish hook-based extensions
• Posts: project notes and updates