GitBot

Product Introduction

1. Definition: GitBot is an open-source, local-first automation platform that transforms AI coding agents (Claude Code, Codex, or OpenCode) into reusable, persistent bots. It operates as a local server, acting as a middleware layer between a web-based UI and the command-line interfaces of these AI agents.
2. Core Value Proposition: GitBot exists to eliminate repetitive prompting and provide controlled, persistent automation for AI-assisted coding tasks. Its core value is enabling developers to create reusable AI bots with defined permissions and persistent conversation threads, allowing them to automate code reviews, refactoring, and maintenance across multiple repositories from a single, local dashboard.

Main Features

1. Reusable Bot Definitions: Users can create and save "bots" with specific instructions, a chosen AI agent (Claude Code, Codex, OpenCode), and a defined permission model. This encapsulates a specific coding task (e.g., "review documentation mismatches") into a reusable asset, eliminating the need to rewrite complex prompts for each new repository or task.
2. Persistent, Thread-Based Conversations: Each execution of a bot is a dedicated "thread" tied to a specific repository folder. This thread maintains the full conversation history with the AI agent, allowing developers to pause, close the browser, and return later to continue the work-in-progress, providing continuity for long-running tasks.
3. Granular Permission & Security Controls: GitBot enforces a security model where bots operate with the user's local permissions. For Claude Code and OpenCode, it can require manual approval for each tool call (file read, write, shell execution). Users can define allowed/disallowed tool lists and choose permission modes (Ask, Auto-approve, Read-only). Crucially, it runs locally with no authentication layer, emphasizing deployment only on trusted networks.
4. Local-First & Agent-Agnostic Architecture: The entire application runs on the user's machine (npm install -g). It does not host AI models itself but interfaces with the locally installed CLIs of supported coding agents. This ensures code never leaves your machine unless sent by the configured AI provider, aligning with privacy and security needs for proprietary codebases.
5. Bot Sharing and Marketplace (Library): Bots can be exported/imported via shareable codes. The project includes a "GitBot Library" marketplace where users can submit bots via pull request. A built-in "Publish to Marketplace" feature guides the AI agent itself in drafting a public listing and preparing the necessary submission files.

Problems Solved

1. Pain Point: Manual, repetitive interaction with AI coding assistants. Developers waste time re-prompting agents for the same type of task (e.g., code review, dependency updates, linting fixes) across different projects.
2. Pain Point: Lack of persistence and context in AI coding sessions. Traditional CLI interactions are ephemeral; losing a session means losing context and history for a complex task.
3. Pain Point: Uncontrolled AI agent permissions. Letting an AI agent run freely in a codebase with write access poses a significant risk of unintended changes.
4. Target Audience: Software engineers, DevOps professionals, and technical leads who regularly use AI coding assistants (Claude Code, Codex, OpenCode) for maintenance, refactoring, and code quality tasks across multiple repositories.
5. Use Cases:
   • Automated Code Review (ShipGuard example): A bot that analyzes a feature branch, compares it against main, and provides a merge/reject recommendation with specific file and line evidence.
   • Documentation Synchronization: A bot that scans a codebase to identify and fix discrepancies between code and its corresponding documentation.
   • Dependency Management: A recurring bot that checks for outdated libraries, suggests updates, and can be instructed to apply patches.
   • Code Standardization: A bot that enforces style guides or architectural patterns across a monorepo.

Unique Advantages

1. Differentiation: Unlike cloud-based AI coding platforms, GitBot is local-first and open-source, giving users full control and visibility. Compared to using raw AI agent CLIs, GitBot adds a crucial layer of reusability, persistence (threads), and permission governance. It is not a hosted service but a local orchestration tool.
2. Key Innovation: Its architecture treats AI agents as runtime engines for higher-level, user-defined "bots." The innovation lies in the abstraction of the prompt/instruction set into a configurable, shareable object with an associated permission model and persistent state (threads). The built-in "Marketplace" prompt for bot creation further leverages AI to help users build and share these abstractions.

Frequently Asked Questions (FAQ)

1. Is GitBot secure for use with private company code? Yes, GitBot's local-first architecture means your source code never leaves your machine to a GitBot server. The AI agent you use (e.g., Claude Code) may send code to its provider per its own configuration, but GitBot itself does not transmit your code. However, it has no built-in authentication, so it must only be run on trusted, private networks.
2. What is the difference between a GitBot "bot" and a simple prompt? A GitBot bot is a persistent, configured entity that bundles a prompt (instructions), a chosen AI agent engine, a permission model, and optional setup steps. Unlike a one-off prompt, a bot is reusable, shareable, and maintains state across executions via threads, turning a static prompt into a managed automation asset.
3. Can I use GitBot with GitHub Copilot or ChatGPT? No, GitBot currently only integrates with Claude Code, Codex, and OpenCode via their official command-line interfaces. It does not support cloud-based chat interfaces like ChatGPT or IDE plugins like GitHub Copilot directly. It is designed for CLI-based AI coding agents.
4. How does GitBot make money if it's free and open-source? GitBot is a community-driven, open-source project (MIT licensed) with no commercial model mentioned. It likely operates on a model of community contribution and may serve as a foundational tool for its creators' other ventures or professional services. There is no telemetry or account requirement.
5. What happens if the AI agent makes a mistake? How do I roll back changes? GitBot does not directly manage version control. It executes tools (like git) through the AI agent. Responsibility for commits and rollbacks remains with the developer. The permission model (especially "Ask before tools") is the primary safeguard, allowing you to review each proposed change. It is recommended to run bots in a clean git state so changes can be easily reviewed and staged.